Systems and methods for camera exposure control

Abstract

Systems and methods for adjusting an exposure parameter of an imaging device are disclosed. A first exposure level of the imaging device is identified, and a first image of a scene is captured via the imaging device at the first exposure level. The first image of the scene comprises a plurality of polarization images corresponding to different degrees and angles of polarization. Each of the polarization images comprise a plurality of color channels. A gradient for the first image is computed based on the plurality of the polarization images, and a second exposure level is computed based on the gradient. A second image of the scene is captured based on the second exposure level, where the gradient of the second image is greater than a gradient for the first image.

Description

FIELD

Aspects of embodiments of the present disclosure relate to the field of digital image processing, and more particularly, to controlling a camera exposure setting for maximizing contrast of an image acquired by the camera.

BACKGROUND

Robotics and other types of computer vision applications may be employed to identify objects in an indoor, industrial setting. The objects to be identified may be homogenous in terms of material, geometry, texture, and color. Identifying and selecting objects may be especially challenging in such a homogenous setting. Accordingly, it is desirable to have a system and method for identifying objects robustly, even when the objects are homogenous objects.

The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure, and therefore, it may contain information that does not form prior art.

SUMMARY

An embodiment of the present disclosure is directed to a method for adjusting an exposure parameter of an imaging device. A first exposure level of the imaging device is identified, and a first image of a scene is captured via the imaging device at the first exposure level. The first image of the scene comprises a plurality of polarization images corresponding to different degrees and angles of polarization. Each of the polarization images comprise a plurality of color channels. A gradient for the first image is computed based on the plurality of the polarization images, and a second exposure level is computed based on the gradient. A second image of the scene is captured based on the second exposure level, where the gradient of the second image is greater than a gradient for the first image.

According to one embodiment, the computing of the gradient for the first image includes: extracting a feature for each of the plurality of color channels for each of the plurality of polarization images; combining the extracted features for generating combined image information; and computing the gradient based on the combined image information.

According to one embodiment, the combining of the extracted features includes invoking a controller for selecting particular ones of the extracted features to be combined.

An embodiment of the present disclosure is also directed to adjusting an exposure parameter of an imaging device, where the method comprises: identifying a first exposure level of the imaging device; capturing a first image of a scene via the imaging device at the first exposure level, the first image of the scene comprising a plurality of polarization images corresponding to different degrees and angles of polarization, each of the polarization images comprising a plurality of color channels; computing a gradient for a first pixel of the first image for each of the plurality of color channels for each of the different degrees and angles of polarization, and outputting a plurality of gradients for the first pixel; selecting a maximum gradient of the plurality of gradients for the first pixel; computing a total gradient score for the first image based on the maximum gradient for the first pixel; computing a second exposure level based on the total gradient score; and capturing a second image of the scene based on the second exposure level, wherein a total gradient score for the second image is greater than the total gradient score for the first image.

According to one embodiment, the first exposure level controls at least one of gain, aperture size, or shutter speed of the imaging device.

According to one embodiment, the first image of the scene further comprises a near infrared image, and the method further comprises: computing a gradient for a first pixel of the first image based on the near infrared image, wherein the plurality of gradients for the first pixel includes the gradient computed based on the near infrared image.

According to one embodiment, the first image of the scene further comprises a non-polarized image, the method further comprising: computing a gradient for the first pixel of the first image based on the non-polarized image, wherein the plurality of gradients for the first pixel includes the gradient computed based on the non-polarized image.

According to one embodiment, method further comprises: computing a gradient for a second pixel of the first image for each of the plurality of color channels for each of the different degrees and angles of polarization, and outputting a plurality of second gradients for the second pixel; and selecting a maximum second gradient of the plurality of second gradients for the second pixel, wherein the computing of the total gradient score for the first image includes computing a sum of the maximum gradient for the first pixel and the maximum second gradient for the second pixel.

According to one embodiment, the method further comprises: comparing the maximum second gradient against a threshold; and discarding the maximum second gradient in response to determining that the maximum second gradient is below the threshold.

According to one embodiment, the method further comprises: comparing the maximum second gradient against a threshold; and accentuating the maximum second gradient in response to determining that the maximum second gradient is above the threshold.

According to one embodiment, the method further comprises: comparing the maximum gradient against the threshold; and accentuating the maximum gradient in response to determining that the maximum gradient is above the threshold.

According to one embodiment, the method further comprises: calculating a difference between the second exposure level and the first exposure level; and in response to the difference being greater than a threshold, iteratively updating a current exposure level of the imaging device, wherein the updating increases a total gradient score of an image captured at the current exposure level.

An embodiment of the present disclosure is further directed to an imaging system that includes an imaging device comprising a polarizing filter, and a processing system coupled to the imaging device. The processing system comprises a processor and memory storing instructions that, when executed by the processor, cause the processor to perform: identifying a first exposure level of the imaging device; capturing a first image of a scene via the imaging device at the first exposure level, the first image of the scene comprising a plurality of polarization images corresponding to different degrees and angles of polarization, each of the polarization images comprising a plurality of color channels; computing a gradient for the first image based on the plurality of the polarization images; computing a second exposure level based on the gradient; and capturing a second image of the scene based on the second exposure level, wherein the gradient of the second image is greater than a gradient for the first image.

An embodiment of the present disclosure is further directed to an imaging system that includes an imaging device comprising a polarizing filter, and a processing system coupled to the imaging device. The processing system comprises a processor and memory storing instructions that, when executed by the processor, cause the processor to perform: identifying a first exposure level of the imaging device; capturing a first image of a scene via the imaging device at the first exposure level, the first image of the scene comprising a plurality of polarization images corresponding to different degrees and angles of polarization, each of the polarization images comprising a plurality of color channels; computing a gradient for a first pixel of the first image for each of the plurality of color channels for each of the different degrees and angles of polarization, and outputting a plurality of gradients for the first pixel; selecting a maximum gradient of the plurality of gradients for the first pixel; computing a total gradient score for the first image based on the maximum gradient for the first pixel; computing a second exposure level based on the total gradient score; and capturing a second image of the scene based on the second exposure level, wherein a total gradient score for the second image is greater than the total gradient score for the first image.

These and other features, aspects and advantages of the embodiments of the present disclosure will be more fully understood when considered with respect to the following detailed description, appended claims, and accompanying drawings. Of course, the actual scope of the invention is defined by the appended claims.

BRIEF DESCRIPTION OF THE DRAWINGS

Non-limiting and non-exhaustive embodiments of the present embodiments are described with reference to the following figures, wherein like reference numerals refer to like parts throughout the various views unless otherwise specified.

FIG. 1A is a block diagram of a system for adjusting an exposure parameter of an imaging device according to one embodiment;

FIG. 1B is a perspective view of a polarization camera module according to one embodiment of the present disclosure;

FIG. 1C is a cross sectional view of a portion of a polarization camera module according to one embodiment of the present disclosure;

FIG. 2 is a flow diagram of a process for iteratively adjusting an exposure level of an imaging device according to one embodiment;

FIG. 3 is a more detailed flow diagram of a process for computing a total gradient or total gradient score of an image according to one embodiment;

FIG. 4 is a conceptual layout diagram of assigning a maximum local gradient to pixel (i, j) based on exemplary local gradient values of the pixel computed for different image modalities according to one embodiment;

FIG. 5A is a conceptual layout diagram of an early fusion technique for fusing multi-modal information according to one embodiment;

FIG. 5B is a conceptual layout diagram of a late fusion technique for fusing multi-modal information according to one embodiment; and

FIG. 5C is a conceptual layout diagram of an algorithm-based technique for fusing multi-modal information according to one embodiment; and

FIG. 6 is a block diagram of an architecture for controlling camera exposure parameters learned through an end-to-end supervised training process according to one embodiment.

DETAILED DESCRIPTION

Hereinafter, example embodiments will be described in more detail with reference to the accompanying drawings, in which like reference numbers refer to like elements throughout. The present disclosure, however, may be embodied in various different forms, and should not be construed as being limited to only the illustrated embodiments herein. Rather, these embodiments are provided as examples so that this disclosure will be thorough and complete, and will fully convey the aspects and features of the present disclosure to those skilled in the art. Accordingly, processes, elements, and techniques that are not necessary to those having ordinary skill in the art for a complete understanding of the aspects and features of the present disclosure may not be described. Unless otherwise noted, like reference numerals denote like elements throughout the attached drawings and the written description, and thus, descriptions thereof may not be repeated. Further, in the drawings, the relative sizes of elements, layers, and regions may be exaggerated and/or simplified for clarity.

Machine vision or computer vision systems process images captured by one or more cameras in order to detect objects in the image. For example, a computer vision application may detect objects in a bin to control a robotic arm to pick an appropriate object from the bin. Robust object detection may depend on the quality of the images that are captured. One factor that may affect the quality of an image is contrast. In general terms, contrast is a difference in color and/or brightness of an object with other objects or the background, that makes the object distinguishable from the other objects or the background. The contrast of an image may depend on the exposure settings of the camera used to capture the image.

Current art mechanisms exist for controlling an exposure level of a camera to capture well-exposed images that are beneficial for machine vision or computer vision algorithms. Some of the current art mechanisms may be catered to natural scenes. In a natural scene, the intrinsic parameters of the scene (e.g. dynamic range, homogeneity, and the like), may vary, but may generally have a wider dynamic range and depict a heterogenous collection of different objects having different colors and luminance. Current art mechanisms that control camera exposure levels for natural scenes may thus not be suitable in indoor, industrial settings. In an industrial setting, the lighting may be poor. The objects to be imaged may also be highly homogeneous in terms of material, geometry, texture, color, and/or the like. Accordingly, it is desirable to determine an exposure level of a camera that will enhance contrast of homogeneous objects in an industrial setting, which may in turn aid object identification by machine vision or computer vision systems.

In general terms, embodiments of the present disclosure are directed to leveraging multi-modal (e.g. multi-dimensional) image data provided by different channels/modalities of image information, for computing a camera exposure level that maximizes image contrast. The computed exposure level may determine one or more exposure settings of the camera including, for example, aperture size, shutter speed (or exposure time), gain, and/or the like. The different channels of image information that may be leveraged for computing the camera exposure level may include, for example, different color polarization channels associated with different angles of linear polarization, near-infrared (NIR) channels, infrared channels, ultraviolet channels, and/or the like.

In one embodiment, one or more cameras set to a current exposure level are used to capture a multi-modal image of the scene in the various channels. The multi-modal image data may be used to compute a gradient for the image. In one embodiment, computing the gradient for the image may entail computing the gradient of each pixel of the image. In general terms, strong gradients improve the detection of keypoints and other features in the image, thereby improving the performance of computer vision systems consuming the image data to perform object instance detection and other tasks.

In one embodiment, the gradient of a pixel in each of the various channels is examined, and the gradient of a channel providing the maximum gradient value that is anticipated to embed maximum pixel data is selected as the gradient value for the pixel across all of the channels. The gradient value may be modulated based on an amount of gradient information that is to be preserved. A gradient map may then be generated based on the modulated gradient values for the various pixels of the image. In one embodiment, the current exposure level of the camera is iteratively adjusted to increase the sum of the gradient values of the gradient map, until a maximum sum is achieved.

In one embodiment, the gradient of a captured multi-modal image is computed by applying a standard gradient computation approach separately to the raw frames captured at each image channel. The standard gradient computation approach may include, for example, Laplacian filter, Difference of Gaussian (DoG) filter, and/or one or more encoder-decoder networks conventional in the art.

In one embodiment, the gradient of a captured multi-modal image is computed using a fusion approach. In this regard, a fusion representation of the multi-modal data may be generated to transfer the multi-modal image into an embedding space. A standard gradient approach may then be used on the embedding space to compute the gradient of the multi-modal image.

In one embodiment, the gradient of the captured multi-modal image is predicted using a deep neural network (e.g. encoder-decoder) architecture. In this regard, the deep neural network may take as an input, the multi-modal image data, and output a corresponding gradient map of the pixels for exposure correction.

In yet some embodiments, the optimized camera exposure parameters are learned by a deep neural network via an end-to-end supervised training process. In this regard, the multi-modal image data may be fed into an encoder for encoding the input data into a fixed-length representation. A decoder network may use the internal representation to output an optimal camera exposure setting for an input multi-modal image. The decoder may be, for example, a fully connected (FC) layer. In some embodiments where spatially varying exposures are desired, the decoder may be a convolutional decoder.

FIG. 1A is a block diagram of a system for adjusting an exposure parameter of an imaging device according to one embodiment. In the embodiment of FIG. 1A, a scene 1 includes various types of objects 2. The objects 2 may be, for example, workpieces placed in a bin with other similar workpieces. In this regard, the objects 2 may be substantially homogenous in terms of material, geometry, texture, and/or color, which may pose difficulties to a machine vision or computer vision system configured to identify and pick an appropriate object from the bin. In some embodiments, the objects 2 include transparent objects.

In one embodiment, the system includes one or more cameras 10, each of the cameras having a lens 12 with a field of view, where the lens 12 and the camera 10 are oriented such that the field of view encompasses the scene 1. The lens 12 is configured to direct light (e.g., focus light) from the scene 1 onto a light sensitive medium such as an image sensor 14 (e.g., a complementary metal oxide semiconductor (CMOS) image sensor or charge-coupled device (CCD) image sensor).

In some embodiments, the one or more cameras 10 may use the same imaging modalities or different imaging modalities. Examples of imaging modalities include, without limitation, monochrome, color, infrared, near-infrared (NIR), ultraviolet, thermal, polarization, and combinations thereof. In one embodiment, the one or more cameras 10 include a polarization camera that uses a polarization imaging modality. In this regard, the polarization camera may be equipped with a polarizer or polarizing filter or polarization mask 16 placed in the optical path between the scene 1 and the image sensor 14. The polarizer or polarization mask 16 may be configured to enable the polarization camera to capture images of the scene 1 with the polarizer set at various specified angles (e.g., at 45° rotations or at 60° rotations or at non-uniformly spaced rotations).

As one example, in the system of FIG. 1A, the polarization mask 16 is a polarization mosaic aligned with the pixel grid of the image sensor 14 in a manner similar to a red-green-blue (RGB) color filter (e.g., a Bayer filter) of a color camera. In a manner similar to how a color filter mosaic filters incoming light based on wavelength such that each pixel in the image sensor 14 receives light in a particular portion of the spectrum (e.g., red, green, or blue) in accordance with the pattern of color filters of the mosaic, a polarization mask 16 using a polarization mosaic filters light based on linear polarization such that different pixels receive light at different angles of linear polarization (e.g., at 0°, 45°, 90°, and 135°, or at 0°, 60° degrees, and 120°). Accordingly, the polarization camera using a polarization mask 16 such as that shown in FIG. 1A is capable of concurrently or simultaneously capturing light at four different linear polarizations. One example of a polarization camera is the Blackfly® S Polarization Camera produced by FLIR® Systems, Inc. of Wilsonville, Oreg.

While the above description relates to some possible implementations of a polarization camera using a polarization mosaic, embodiments of the present disclosure are not limited thereto, and encompass other types of polarization cameras that are capable of capturing images at multiple different polarizations. For example, the polarization filter 16 may have fewer than or more than four different polarizations, or may have polarizations at different angles (e.g., at angles of polarization of: 0°, 60° degrees, and 120° or at angles of polarization of 0°, 30°, 60°, 90°, 120°, and 150°). As another example, the polarization filter 16 may be implemented using an electronically controlled polarization filter, such as an electro-optic modulator (e.g., may include a liquid crystal layer), where the polarization angles of the individual pixels of the filter may be independently controlled, such that different portions of the image sensor 14 receive light having different polarizations. Furthermore, while the above examples relate to the use of a linear polarizing filter, embodiments of the present disclosure are not limited thereto, and also include the use of polarization cameras that include circular polarizing filters (e.g., linear polarizing filters with a quarter wave plate). Accordingly, in some embodiments of the present disclosure, a polarization camera uses a polarizing filter to capture multiple polarization raw frames at different polarizations of light, such as different linear polarization angles and different circular polarizations (e.g., handedness).

FIG. 1B is a perspective view of a polarization camera module 10′ according to one embodiment of the present disclosure. FIG. 1C is a cross sectional view of a portion of a polarization camera module 10A′ according to one embodiment of the present disclosure. Some aspects of embodiments of the present disclosure relate to a polarization camera module in which multiple polarization cameras (e.g., multiple cameras, where each camera has a polarizing filter in its optical path) are arranged adjacent to one another and in an array and may be controlled to capture images in a group (e.g., a single trigger may be used to control all of the cameras in the system to capture images concurrently or substantially simultaneously). The polarizing filters in the optical paths of each of the cameras in the array cause differently polarized light to reach the image sensors of the cameras. The individual polarization cameras in the camera system have optical axes that are substantially perpendicular to one another, are placed adjacent to one another, and have substantially the same field of view, such that the cameras in the camera system capture substantially the same view of a scene 1, but with different polarizations. In some embodiments, the individual polarization cameras are arranged such that parallax shift between cameras is substantially negligible based on the designed operating distance of the camera system to objects in the scene, where larger spacings between the cameras may be tolerated if the designed operating distance is large.

For example, in the embodiment of the polarization camera module 10′ shown in FIG. 1B, four cameras 10A′, 10B′, 10C′, and 10D′ are arranged in a 2×2 grid to form a camera array, where the four cameras have substantially parallel optical axes. The four cameras may be controlled together such that they capture images substantially simultaneously and using the same exposure settings (e.g., same aperture, length of exposure, and gain or “ISO” settings). In various embodiments of the present disclosure, each of the separate cameras 10A′, 10B′, 10C′, and 10D′ includes a different polarizing filter.

In some embodiments, three of the cameras 10A′, 10B′, 10C′ of the camera array are polarization cameras with a different polarizing filter set at a different angle of polarization (e.g. 0 degrees, 60 degrees, and 120 degrees). The fourth camera 10D′ of the array may equipped with a near-infrared (NIR) sensor for perceiving light in a near-infrared range. The fourth camera 10D′ may or may not include a polarizing filter.

FIG. 1C shows a cross sectional view of two of the polarization cameras 10A′ and 10B′ shown in FIG. 1B. As seen in FIG. 1C, each a polarization camera (10A′ and 10B′) system includes a corresponding lens, a corresponding image sensor, and a corresponding polarizing filter. In particular, polarization camera 10A′ includes lens 12A′, image sensor 14A′, and polarizing filter 16A′. Likewise, polarization camera 10B′ includes lens 12B′, image sensor 14B′, and polarizing filter 16B′. In some embodiments of the present disclosure, the image sensors of four cameras 10A′, 10B′, 10C′, and 10D′ are monolithically formed on a same semiconductor die, and the four cameras are located in a same housing with separate apertures for the lenses 12 corresponding to the different image sensors. Similarly, the polarizing filters 16 may correspond to different portions of a single physical layer that has different polarizing filters (e.g., different linear polarizing angles) in different regions of the layer (corresponding to the different cameras).

In some embodiments of the present disclosure, each of the cameras in the camera system 10′ has a corresponding polarizing filter that is configured to filter differently polarized light. For example, in the embodiment shown in FIG. 1C, polarizing filter 16A′ of camera 10A′ may be a linear polarizing filter oriented at an angle of 0°, polarizing filter 16B′ of camera 10B′ may be a linear polarizing filter oriented at an angle of 45°, polarizing filter 16C′ of camera 10C′ may be a linear polarizing filter oriented at an angle of 90°, and polarizing filter 16D′ of camera 10D′ may be a linear polarizing filter oriented at an angle of 135°.

In the embodiment where only three of the cameras in the array are polarization cameras, polarizing filter 16A′ of camera 10A′ may be a linear polarizing filter oriented at an angle of 0°, polarizing filter 16B′ of camera 10B′ may be a linear polarizing filter oriented at an angle of 60°, and polarizing filter 16C′ of camera 10C′ may be a linear polarizing filter oriented at an angle of 120°. In some embodiments, one or more of the cameras may include a circular polarizer. In some embodiments of the present disclosure, the camera system 10′ includes polarizing filters configured to filter light in at least two different polarizations. In some embodiments of the present disclosure, the camera system 10′ includes polarizing filters configured to filter light in at least three different polarizations. In the embodiment shown in FIG. 1C, the polarizing filter 16 is located behind the lens 12 (e.g., between the lens 12 and the image sensor 14), but embodiments of the present disclosure are not limited thereto. In some embodiments, the polarizing filter is located in front of the lens 12.

In some embodiments, the various individual cameras of the camera array are registered with one another by determining their relative poses (or relative positions and orientations) by capturing multiple images of a calibration target, such as a checkerboard pattern, an ArUco target (see, e.g., Garrido-Jurado, Sergio, et al. “Automatic generation and detection of highly reliable fiducial markers under occlusion.” Pattern Recognition 47.6 (2014): 2280-2292.) or a ChArUco target (see, e.g., An, Gwon Hwan, et al. “Charuco board-based omnidirectional camera calibration method.” Electronics 7.12 (2018): 421.). In particular, the process of calibrating the targets may include computing intrinsic matrices characterizing the internal parameters of each camera (e.g., matrices characterizing the focal length, image sensor format, and principal point of the camera) and extrinsic matrices characterizing the pose of each camera with respect to world coordinates (e.g., matrices for performing transformations between camera coordinate space and world or scene coordinate space).

While not shown in FIG. 1C, in some embodiments of the present disclosure, each polarization camera may also include a color filter having in a mosaic pattern such as a Bayer filter, such that individual pixels of the image sensors 14 receive light corresponding to, for example, red (R), green (G), and blue (B) portions of the spectrum, such that each camera captures light in a visible portion of the electromagnetic spectrum in accordance with a mosaic pattern. For example, three of the cameras 10A′, 10B′, and 10C′ may be color polarization cameras where each of the color polarization cameras may generate a polarization image having R, G, and B color components. In some embodiments, a demosaicing process is used to compute separate red, green, and blue channels from the raw data. In some embodiments of the present disclosure, each polarization camera may be used without a color filter or with filters used to transmit or selectively transmit various other portions of the electromagnetic spectrum, such as infrared light.

As a result, the one or more cameras 10 in FIG. 1A capture multiple input images/raw frames 18 (e.g. polarization raw frames, NIR raw frames, non-polarized color raw frames, etc.) of the scene 1. Each of the polarization raw frames may correspond to an image taken behind a polarization filter at a different angle of polarization ϕ_pol (e.g., 0 degrees, 60 degrees, or 120 degrees). In one embodiment, each of the raw frames 18 is captured from substantially the same pose with respect to the scene 1, as opposed to capturing the polarization raw frames from disparate locations and orientations with respect to the scene. For example, the images captured via polarization camera(s) with the polarization filter set at 0 degrees, 60 degrees, and 120 degrees, and the image capture by the NIR camera, are all captured at a same location and orientation. Thus, according to one embodiment, a particular point of an image (referred to as an image pixel), is depicted by corresponding pixels in the polarization raw frames and the NIR raw frame.

In one embodiment, one or more of the cameras 10 (e.g. the polarization cameras) include a color filter 17 having a mosaic pattern such as, for example, a Bayer filter. In this regard, the raw frames 18 generated by the camera may be color polarized raw frames. In one embodiment, the color filter 17 may allow individual pixels of the image sensors 14 to receive light corresponding to, for example, red (R), green (G), and blue (B) portions of the spectrum, such that each camera captures light in a visible portion of the electromagnetic spectrum in accordance with a mosaic pattern. In some embodiments, a demosaicing process is used to compute separate red, green, and blue channels from the raw polarization data.

In one embodiment, the system of FIG. 1A further includes a processing circuit 100 configured to calculate the gradient of a multi-modal image, and automatically adjust an exposure parameter of the cameras 10 based on the calculated gradient for maximizing contrast of the images obtained by the cameras. In this regard, the raw image frames 18 generated by the one or more cameras 10 are provided to the processing circuit 100 for computing the gradients of the various images. In one embodiment, the raw frames include polarization raw frames that are used by the processing circuit 100 to calculate a degree of linear polarization (DOLP) and an angle of linear polarization (AOLP) for each pixel of an image of the scene 1, for each of the various channels, including R, G, and B channels. In some embodiments where a polarizing filter is used with an NIR camera, the DOLP and AOLP may also be computed for the NIR channels.

The degree of linear polarization DOLP may be calculated using Stokes parameters as follows:

$\mathrm{DOLP}=\frac{\sqrt{Q^2+U^2}}{I}=\frac{I_{\mathrm{pol}}}{I}$ where I_pol is the fraction of light that is polarized and is thus the polarized radiance, I is the total radiance incident on the surface of the object, Q is the difference in the measured polarization intensity between the vertical and horizontal polarization states, and U is the difference in the measured polarization intensity between the +45 deg and −45 deg polarization states.

The angle of linear polarization AOLP may be calculated as follows:

$\mathrm{AOLP}=\frac{1}{2}{\tan }^{-1}\left(\frac{U}{Q}\right)$

In one embodiment, the processing circuit 100 is configured to compute a gradient of the DOLP and AOLP values associated with a pixel for each color channel, for each of the different angles of linear polarization, and select a maximum gradient for assigning to the pixel. For example, the maximum gradient for a first pixel may be the gradient of the AOLP values in the red channel, while the maximum gradient for a second pixel may be the gradient of the DOLP values in the blue channel. A gradient map of the image may be generated based on the maximum gradient values.

In one embodiment, the processing circuit 100 applies a standard gradient computation approach separately to each of the image modalities for computing the gradient values of the multi-modal image. The standard gradient computation approach may include, for example, Laplacian filter, Difference of Gaussian (DoG) filter, and/or one or more encoder-decoder networks conventional in the art.

In some embodiments, one or more deep neural networks may be invoked for computing the gradient of the multi-modal image. For example, the processing circuit 100 may invoke the one or more deep neural networks to extract features of the image in the different image modalities, and combine the extracted features for generating combined image information. For example, the processing circuit 100 may invoke the one or more deep neural networks to extract a feature map of the image for each of the plurality of color channels and for each of the plurality of polarization raw images, and combine the extracted features for generating the combined image information. In one embodiment, the extracted features are ones that are predicted to provide a maximum gradient score for the image. The processing circuit 100 may then provide the combined image information to an encoder-decoder architecture for predicting the gradient for the image. In this regard, the extracted features may be decoded into gradient values. The gradient values may then be combined to compute a final gradient score for the image.

In one embodiment, the processing circuit 100 computes a measure or score regarding the total gradient for the multi-modal image based on the maximum gradients in the gradient map, and iteratively adjusts the exposure level of the one or more cameras 10 for increasing the gradient of the image until a maximum gradient level or gradient score is reached. The processing circuit 100 may transmit a signal 102 to the one or more cameras 10 for modifying one or more exposure parameters of the camera based on the modified exposure level. For example, the camera may adjust at least one of aperture size, shutter speed, and/or gain, based on the modified exposure level. In one embodiment, the image captured according to the second exposure setting has a total gradient that is greater than the total gradient of the image captured according to the first exposure setting.

FIG. 2 is a flow diagram of a process for iteratively adjusting an exposure level of an imaging device according to one embodiment. It should be understood that the sequence of steps of the process is not fixed, but can be modified, changed in order, performed differently, performed sequentially, concurrently, or simultaneously, or altered into any desired sequence, as recognized by a person of skill in the art.

At block 200, the processing circuit 100 obtains a current exposure level, E_t, for the one or more cameras 10, and sets one or more exposure parameters of the camera (e.g. aperture size, shutter speed, and/or gain), based on the obtained exposure setting. For example, the processing circuit 100 may access an entry in an exposure table based on the identified exposure level, and retrieve values in the entry corresponding to one or more of aperture size, shutter speed, and/or gain.

At block 202, the processing circuit 100 invokes the one or more cameras 10 to capture one or more images of the scene 1. In this regard, the one or more cameras may capture multiple raw frames 18 of the scene 1 using the current exposure parameters. Each of the multiple raw frames 18 may provide image data of the scene 1 in a different image modality. For example, the raw frames 18 may include color polarization raw frames, NIR raw frames, and/or the like.

At block 204, the processing circuit 100 computes a total amount of gradient (GradMag) of the image based on the gradients computed for each pixel of the image using the multi-modal image data.

At block 206, the processing circuit 100 computes a new exposure level, E_t+1 based on the current exposure value (E_t) and the computed total gradient (GradMag) of the image, according to the following formula:E_t+1=1+K*(1-GradMag)*E_t where K is a proportionality constant in the range of [0, 1] that provides a balance between convergence speed and the stability of convergence. For example, a high K value may allow faster convergence, but be prone to oscillation and overshooting. In one embodiment, the processing circuit 100 may set K at runtime and further tune K based on the scene. In one embodiment, the processing circuit may select K based on a type of scene that is to be captured.

At block 208, the processing circuit 100 sets the camera exposure of the one or more cameras 10 to the new exposure level E_t+1.

At block 210, a determination is made as to whether a difference between the new exposure level E_t+1 and the previous exposure level E_t is greater than a threshold value. If the answer is YES, the process continues to compute the total gradient amount of the new image taken at the new exposure level E_t+1.

If, however, the difference between the new exposure level E_t+1 and the previous exposure level E_t is not greater than the threshold, a maximum gradient is deemed to have been achieved, and the process stops.

FIG. 3 is a more detailed flow diagram of the process at block 204 of computing a total gradient or total gradient score (GradMag) of an image according to one embodiment. At block 300, the processing circuit 100 computes a gradient for each pixel for each image channel (referred to as a multimodal gradient). In one embodiment, the various image modalities/channels for which the gradient is computed include the various color channels of the raw polarization frames. In this regard, an AOLP and DOLP value may be computed for each pixel of the image for each of the R, G, and B color channels, using Stokes parameters based on three of the color polarized raw frames, and a gradient may be computed for each AOLP and DOLP for each of the R, G, and B color channels. In some embodiments, a gradient is computed for non-polarization channels such as, for example, an NIR channel, non-polarized color channels, and/or the like. In one embodiment, the gradient of a pixel at each channel may be computed using a standard gradient approach including, for example, Laplacian filter, Difference of Gaussian (DoG) filter, and/or one or more encoder-decoder networks conventional in the art.

At block 302, the processing circuit 100 selects a maximum gradient value for each pixel of the image. A gradient map of the image may be generated based on the selected maximum gradients. In one embodiment, a maximum gradient is selected from the gradients of the AOLP and DOLP computed for each of the R, G, and B color channels as follows:g(i,j)=argmax_{AOLP(r,g,b),DOLP(r,g,b)}f(grad(AOLP(p_r,g,b(i,j),DOLP(p_r,g,b(i,j)))where g(i,j) is a local gradient value at pixel p(i,j), grad( ) is gradient function which takes an image as input, and f( ) is a linear or non-linear function (e.g. a logarithmic function) that adjusts the raw gradient value.

In some embodiments, the processing circuit 100 selects a maximum of the gradients of pixel values in non-polarization channels such as the NIR channel (n) or non-polarized color channels (r, g, b) as follows:g(i,j)=argmax_r,g,b,nf(grad(p_r,g,b,n(i,j)))

In some embodiments, the processing circuit 100 considers the gradient computations of both the polarization and non-polarization channels, and selects a maximum of the gradients for assigning to the corresponding pixel of the gradient map.

At block 304, the processing circuit 100 computes a total amount of gradient (GradMag) for the image as a sum of modulated local maximum gradients g(i, j) identified for each pixel as follows:GradMag=SUM(alpha*(g(i,j)−delta))/N where,N=alpha*(1−delta)

In one embodiment, N is a normalization factor that allows GradMag to take a value between 0 and 1. Alpha and delta may be control parameters for modulating the amount of gradient information that is to be preserved. For example, the processing circuit 100 may select a small alpha value to accentuate strong gradient information. The processing circuit 100 may also select a large alpha value to emphasize subtle gradient information. Delta may be a threshold value preset by the processing circuit 100 for filtering out gradient values that may be deemed to be noise. In this regard, a minimum value of g(i,j) computed by the processing circuit 100 may be equal to delta.

FIG. 4 is a conceptual layout diagram of assigning a maximum local gradient 400 to pixel (i, j) based on exemplary local gradient values of the pixel computed for different image modalities according to one embodiment. In the example of FIG. 4, the computed gradient of the AOLP for the red, green, and blue channels 402, 404, 406 are respectively be 3, 6, and 2. The gradient of the DOLP of the red, green, and blue channels 408, 410, 412 are respectively be 1, 0, and 7. In addition, the gradient of the pixel in the NIR channel 414 is 4. In the example of FIG. 4, the maximum gradient for pixel (i,j) is the gradient of the DOLP in the blue channel 412. Thus, the gradient map g, at pixel (i,j), is assigned a gradient value of 7.

In some embodiments, deep learning techniques may be employed for computing a gradient of a multi-modal image. For example, multimodal data fusion techniques, such as early fusion and late fusion, may be employed to extract and fuse information from each image modality. Standard gradient computation techniques may then be employed on the fused representation of the multi-modal image for computing the gradient of the multi-modal image.

In some embodiments, features of the various modalities of the multi-modal image may be extracted, and a controller unit may be employed to determine whether features are to be combined, and if so, which of the features are to be fused. The features may be extracted via one or more layers of a convolutional neural network. In one embodiment, the controller unit may be configured to perform a one-layer or multi-layer fusion. The output of the controller unit may then pass through a neural architecture for generating a gradient prediction for the multi-modal image.

FIG. 5A is a conceptual layout diagram of an early fusion technique for fusing multi-modal information according to one embodiment. In one embodiment, the processing circuit 100 identifies data of a multi-modal image in different imaging modalities 500, 502, and performs data-level concatenation of the extracted features. In one embodiment, the processing circuit 100 invokes a fusion module 504 for performing the data-level concatenation. In this regard, the fusion module 504 may execute an algorithm for the data-level concatenation such as, for example, principal component analysis (PCA), canonical correlation analysis, independent component analysis, or the like. The fused image information may then be provided to an encoder-decoder architecture 506 for computing a gradient 508 of the multi-modal image according to standard approaches.

FIG. 5B is a conceptual layout diagram of a late fusion technique for fusing multi-modal information according to one embodiment. In one embodiment, the processing circuit 100 extracts data features of the multi-modal image in different imaging modalities 500′, 502′ via feature encoders 550, 552. The extracted features may then be combined using late fusion approaches 506′ such as, for example, Bayes rules, max-fusion, and average-fusion. The fused image information may be provided to an encoder-decoder architecture 506′ for computing a gradient 508′ of the multi-modal image according to standard approaches.

FIG. 5C is a conceptual layout diagram of an algorithm-based technique for fusing multi-modal information according to one embodiment. In one embodiment, the processing circuit 100 extracts data features of the multi-modal image in different imaging modalities 500″, 502″ via feature encoders 560, 562. The extracted features may be provided to a fusion controller 564 to determine whether features are to be combined, and if so, which of the features are to be fused. In this regard, the fusion controller 564 may be configured to search combinations of different convolutional layers to fuse, and select a combination (e.g. one-layer or multi-layer) that provides a desired performance. For example, the fusion controller 564 may be optimized for overall task performance. The task to be optimized may be, for example, accurate image detection. In one embodiment, the fused image information may be provided to an encoder-decoder architecture 506″ for computing a gradient 508″ of the multi-modal image according to standard approaches.

FIG. 6 is a block diagram of an architecture for controlling camera exposure parameters learned through an end-to-end supervised training process according to one embodiment. Raw input images 600 captured by the one or more cameras 10 in the different imaging modalities are provided to an encoder 602 such as, for example, a convolutional neural network. In the embodiments where the images include polarized images, the neural network may be implemented as a Polarized CNN backbone as described in PCT Patent Application No. US2020/048604, the content of which is incorporated herein by reference.

In one embodiment, the encoder 602 is configured to generate feature maps based on the input images 600. The generated feature maps may be provided to a decoder 604 for outputting an optimal camera exposure value based on the captured images. The computed camera exposure value may then be supplied to the one or more cameras 10 (or other cameras imaging, for example, the same scene or scenes under similar conditions) to control the exposure settings when capturing subsequent images.

According to various embodiments of the present disclosure, the processing circuit 100 is implemented using one or more electronic circuits configured to perform various operations as described in more detail below. Types of electronic circuits may include a central processing unit (CPU), a graphics processing unit (GPU), an artificial intelligence (AI) accelerator (e.g., a vector processor, which may include vector arithmetic logic units configured efficiently perform operations common to neural networks, such dot products and softmax), a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a digital signal processor (DSP), or the like. For example, in some circumstances, aspects of embodiments of the present disclosure are implemented in program instructions that are stored in a non-volatile computer readable memory where, when executed by the electronic circuit (e.g., a CPU, a GPU, an AI accelerator, or combinations thereof), perform the operations described herein to compute a segmentation map 20 from input polarization raw frames 18. The operations performed by the processing circuit 100 may be performed by a single electronic circuit (e.g., a single CPU, a single GPU, or the like) or may be allocated between multiple electronic circuits (e.g., multiple GPUs or a CPU in conjunction with a GPU). The multiple electronic circuits may be local to one another (e.g., located on a same die, located within a same package, or located within a same embedded device or computer system) and/or may be remote from one other (e.g., in communication over a network such as a local personal area network such as Bluetooth®, over a local area network such as a local wired and/or wireless network, and/or over wide area network such as the internet, such a case where some operations are performed locally and other operations are performed on a server hosted by a cloud computing service). One or more electronic circuits operating to implement the processing circuit 100 may be referred to herein as a computer or a computer system, which may include memory storing instructions that, when executed by the one or more electronic circuits, implement the systems and methods described herein.

While the present invention has been described in connection with certain exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims, and equivalents thereof.

Claims

1. A method for adjusting an exposure parameter of an imaging device, the method comprising: identifying a first exposure level of the imaging device;capturing a first image of a scene via the imaging device at the first exposure level, the first image of the scene comprising a plurality of polarization images corresponding to different degrees and angles of polarization, each of the polarization images comprising a plurality of color channels;computing a gradient for the first image based on the plurality of the polarization images;computing a second exposure level based on the gradient; andcapturing a second image of the scene based on the second exposure level, wherein the gradient of the second image is greater than a gradient for the first image.

2. The method of claim 1, wherein the computing of the gradient for the first image includes: extracting a feature for each of the plurality of color channels for each of the plurality of polarization images;combining the extracted features for generating combined image information; andcomputing the gradient based on the combined image information.

3. The method of claim 1, wherein the combining of the extracted features includes invoking a controller for selecting particular ones of the extracted features to be combined.

4. The method of claim 1, wherein the first exposure level controls at least one of gain, aperture size, or shutter speed of the imaging device.

5. The method of claim 1, wherein the first image of the scene further comprises a near infrared image, the method further comprising: computing a gradient for a first pixel of the first image based on the near infrared image, wherein the plurality of gradients for the first pixel includes the gradient computed based on the near infrared image.

6. The method of claim 1, wherein the first image of the scene further comprises a non-polarized image, the method further comprising: computing a gradient for the first pixel of the first image based on the non-polarized image, wherein the plurality of gradients for the first pixel includes the gradient computed based on the non-polarized image.

7. The method of claim 1 further comprising: computing a gradient for a second pixel of the first image for each of the plurality of color channels for each of the different degrees and angles of polarization, and outputting a plurality of second gradients for the second pixel; andselecting a maximum second gradient of the plurality of second gradients for the second pixel, wherein the computing of the total gradient score for the first image includes computing a sum of the maximum gradient for the first pixel and the maximum second gradient for the second pixel.

8. The method of claim 7 further comprising: comparing the maximum second gradient against a threshold; anddiscarding the maximum second gradient in response to determining that the maximum second gradient is below the threshold.

9. The method of claim 7 further comprising: comparing the maximum second gradient against a threshold; andaccentuating the maximum second gradient in response to determining that the maximum second gradient is above the threshold.

10. The method of claim 9 further comprising: comparing the maximum gradient against the threshold; andaccentuating the maximum gradient in response to determining that the maximum gradient is above the threshold.

11. The method of claim 1 further comprising: calculating a difference between the second exposure level and the first exposure level; andin response to the difference being greater than a threshold, iteratively updating a current exposure level of the imaging device, wherein the updating increases a total gradient score of an image captured at the current exposure level.

12. A method for adjusting an exposure parameter of an imaging device, the method comprising: identifying a first exposure level of the imaging device;capturing a first image of a scene via the imaging device at the first exposure level, the first image of the scene comprising a plurality of polarization images corresponding to different degrees and angles of polarization, each of the polarization images comprising a plurality of color channels;computing a gradient for a first pixel of the first image for each of the plurality of color channels for each of the different degrees and angles of polarization, and outputting a plurality of gradients for the first pixel;selecting a maximum gradient of the plurality of gradients for the first pixel;computing a total gradient score for the first image based on the maximum gradient for the first pixel;computing a second exposure level based on the total gradient score; andcapturing a second image of the scene based on the second exposure level, wherein a total gradient score for the second image is greater than the total gradient score for the first image.

13. An imaging system comprising: an imaging device comprising a polarizing filter; anda processing system coupled to the imaging device, the processing system comprising a processor and memory storing instructions that, when executed by the processor, cause the processor to perform: identifying a first exposure level of the imaging device;capturing a first image of a scene via the imaging device at the first exposure level, the first image of the scene comprising a plurality of polarization images corresponding to different degrees and angles of polarization, each of the polarization images comprising a plurality of color channels;computing a gradient for the first image based on the plurality of the polarization images;computing a second exposure level based on the gradient; andcapturing a second image of the scene based on the second exposure level, wherein the gradient of the second image is greater than a gradient for the first image.

14. The system of claim 13, wherein the instructions that cause the processor to perform computing of the gradient for the first image include instructions that cause the processor to perform: extracting a feature for each of the plurality of color channels for each of the plurality of polarization images;combining the extracted features for generating combined image information; andcomputing the gradient based on the combined image information.

15. The system of claim 13, wherein the instructions that cause the processor to perform combining of the extracted features include instructions that cause the processor to perform invoking a controller for selecting particular ones of the extracted features to be combined.

16. An imaging system comprising: an imaging device comprising a polarizing filter; anda processing system coupled to the imaging device, the processing system comprising a processor and memory storing instructions that, when executed by the processor, cause the processor to perform: identifying a first exposure level of the imaging device;capturing a first image of a scene via the imaging device at the first exposure level, the first image of the scene comprising a plurality of polarization images corresponding to different degrees and angles of polarization, each of the polarization images comprising a plurality of color channels;computing a gradient for a first pixel of the first image for each of the plurality of color channels for each of the different degrees and angles of polarization, and outputting a plurality of gradients for the first pixel;selecting a maximum gradient of the plurality of gradients for the first pixel;computing a total gradient score for the first image based on the maximum gradient for the first pixel;computing a second exposure level based on the total gradient score; andcapturing a second image of the scene based on the second exposure level, wherein a total gradient score for the second image is greater than the total gradient score for the first image.

17. The system of claim 16, wherein the first exposure level controls at least one of gain, aperture size, or shutter speed of the imaging device.

18. The system of claim 16, wherein the first image of the scene further comprises a near infrared image, wherein the instructions further cause the processor to perform: computing a gradient for the first pixel of the first image based on the near infrared image, wherein the plurality of gradients for the first pixel includes the gradient computed based on the near infrared image.

19. The system of claim 16, wherein the first image of the scene further comprises a non-polarized image, wherein the instructions further cause the processor to perform: computing a gradient for the first pixel of the first image based on the non-polarized image, wherein the plurality of gradients for the first pixel includes the gradient computed based on the non-polarized image.

20. The system of claim 16, wherein the instructions further cause the processor to perform: computing a gradient for a second pixel of the first image for each of the plurality of color channels for each of the different degrees and angles of polarization, and outputting a plurality of second gradients for the second pixel; andselecting a maximum second gradient of the plurality of second gradients for the second pixel, wherein the computing of the total gradient score for the first image includes computing a sum of the maximum gradient for the first pixel and the maximum second gradient for the second pixel.

21. The system of claim 20, wherein the instructions further cause the processor to perform: comparing the maximum second gradient against a threshold; anddiscarding the maximum second gradient in response to determining that the maximum second gradient is below the threshold.

22. The system of claim 20, wherein the instructions further cause the processor to perform: comparing the maximum second gradient against a threshold; andaccentuating the maximum second gradient in response to determining that the maximum second gradient is above the threshold.

23. The system of claim 22, wherein the instructions further cause the processor to perform: comparing the maximum gradient against the threshold; andaccentuating the maximum gradient in response to determining that the maximum gradient is above the threshold.

24. The system of claim 16, wherein the instructions further cause the processor to perform: calculating a difference between the second exposure level and the first exposure level; andin response to the difference being greater than a threshold, iteratively updating a current exposure level of the imaging device, wherein the updating increases a total gradient score of an image captured at the current exposure level.