Images for perception modules of autonomous vehicles

Abstract

Disclosed are devices, systems and methods for processing an image. In one aspect a method includes receiving an image from a sensor array including an x-y array of pixels, each pixel in the x-y array of pixels having a value selected from one of three primary colors, based on a corresponding x-y value in a mask pattern. The method may further include generating a preprocessed image by performing preprocessing on the image. The method may further include performing perception on the preprocessed image to determine one or more outlines of physical objects.

Description

TECHNICAL FIELD

This document relates to reducing the data complexity for analysis and bandwidth required of autonomous vehicle images.

BACKGROUND

Autonomous vehicle navigation is a technology for sensing the position and movement of a vehicle, and based on the sensing, autonomously controlling the vehicle to navigate towards a destination. Autonomous vehicle navigation can have important applications in transportation of people, goods and services. One of the components of autonomous driving, which ensures the safety of the vehicle and its passengers, as well as people and property in the vicinity of the vehicle, is analysis of images taken from vehicle cameras. The images may be used to determine fixed or moving obstacles in the path of autonomous vehicle.

SUMMARY

Disclosed are devices, systems and methods for processing images of an area surrounding a vehicle. In some embodiments, light detection and ranging (LiDAR) sensors may be used to acquire the images based on reflections captured from the surrounding area. In one aspect, a method for processing an image taken from an autonomous vehicle is disclosed. The method includes receiving a raw image from a camera, the image including three values for each of three primary colors, and/or selecting one of the three values for each pixel in the image and discarding the other two values, wherein the selecting is performed in a pattern. The method may further include performing preprocessing on the reduced image, and/or performing perception on the preprocessed image to determine one or more outlines of physical objects in a vicinity of the autonomous vehicle.

The method may further include the following features in any combination. The selecting may reduce a data size of the raw image by a factor of ⅔. The pattern may be a Bayer pattern. The Bayer pattern may be a red-green-green-blue pattern assigned to a 2-pixel by 2-pixel array repeated across the raw image. The pattern may include a greater number of green pixel values than both red and blue. The pattern may be selected such that a value of every other pixel along a row in the reduced image corresponds to green value of the raw image. The pattern may be selected such that a value of every other pixel along a column in the reduced image corresponds to green value of the raw image. The generating the reduced image may be performed using one or more color-selective filters. Each x-y value in the pattern may be from one of three possible values. The preprocessing may be performed on the image from the sensor array without human perception image enhancement. The human perception image enhancement may include one or more of de-mosaicing, white balancing, and noise reduction. The preprocessing may not include scaling one or more pixels' R, G, or B value for white balancing. The preprocessing may not include reconstruction a full color image from incomplete color samples output from the sensor array overlaid with a color filter array for de-mosaicing. The preprocessing may not include noise reduction, wherein noise reduction includes reduction of salt and pepper noise, wherein a noisy pixel bears little relation to the color of surrounding pixels, or reduction of Gaussian noise. The preprocessing may include image cropping. The preprocessing may include image resizing. The preprocessing may include image compression. The sensor array may be a camera.

In another aspect, the above-described method is embodied in the form of executable code stored in a computer-readable program medium.

In yet another aspect, a device that is configured or operable to perform the above-described method is disclosed. The device may include a processor that is programmed to implement this method.

The above and other aspects and features of the disclosed technology are described in greater detail in the drawings, the description and the claims.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 depicts a flowchart for processing an image for use by an autonomous vehicle, in accordance with some example embodiments;

FIG. 2 depicts an example of reducing the data size a color image for use by an autonomous vehicle, in accordance with some example embodiments;

FIG. 3A depicts examples of Bayer images at different zooms to show the patterns;

FIG. 3B depicts an example of a Bayer pattern image in greyscale;

FIG. 4 depicts an example of a contrast image and a perception result, in accordance with some example embodiments; and

FIG. 5 depicts an example of an apparatus, in accordance with some example embodiments.

DETAILED DESCRIPTION

Pictures taken by still or video cameras are typically intended for human viewing. The pictures are in full color with high resolution in three primary color such as red (R), green (G) and blue (B). Image processing including vision tasks such as object detection, semantic segmentation, and others typically use processed, well rendered images that are intended for human eyes. However, images for use by machines do not need to have the same characteristics as images intended for human viewing. For example, de-mosaicing, white balancing, color reduction, and other image processing may not be necessary for machine vision used for autonomous vehicles. Since the foregoing image processing tasks do not add additional information, they may not be needed or be useful for machine images. Not performing some processing tasks such as white balancing, which can cause over exposure, also improves the images. Moreover, the amount of data required to represent an image for machine use is less than the full color RGB representation of an image.

In some example embodiments, the RGB information for each pixel in an image may be reduced so that instead of each pixel having intensity values for each of R, G, and B, each pixel has only one intensity corresponding to R, G or B. A predetermined pattern of R, G, and B may be assigned to the array of pixels in an image. By reducing the number of intensity values per pixel from three to one, the data required to represent the image is reduced to ⅓ of the data needed for a full color RGB image. In this way, the amount of data needed to represent an image is reduced to ⅓ while maintaining color sensitivity needed for colored objects in the image. Reducing the amount of data needed to represent the image reduces the bandwidth needed to transfer the image in a fixed amount of time or allows the image to be transferred or processed in less time. Both of these improve machine vision performance and responsiveness. As an illustrative example, a camera image with a size of 200 pixels by 300 pixels has 60,000 pixels. If each pixel is represented by an 8-bit red (R) value, an 8-bit green (G) value, and an 8-bit blue (B) value, the total number of bits required to represent the color image is 300 (pixels)×200 (pixels)×8 (bits)×3 (colors)=1,440,000 bits. By applying the pattern to select one of R, G, or B, for each pixel in a pattern, the number of bits needed to represent the image is reduced to 480,000. The foregoing is an example for illustrative purposes. Other numbers of pixels per image or bits of resolution per color can also be used.

FIG. 1 depicts a flowchart 100 for processing an image for use by an autonomous vehicle, in accordance with some example embodiments. At 110, an image is received. At 120, the data required to represent the raw image is reduced. At 130, the reduced image is preprocessed. At 140, perception is performed on the preprocessed image. At 150, the perception result is provided as an output.

At 110, an image is received form a camera or LiDAR or other image generating device. For example, an image from a solid-state camera such as a charge coupled device (CCD) camera is received. The camera may have separate outputs for R, G, and B or may have a composite output. As an example, R, G, and B may each be represented by 8-bit luminance values or may be represented by analog voltages. In another example, the image may be from a multi-spectral Light Detection and ranging (LiDAR) sensor. Each “color” may be represented by an 8-bit or another bit resolution value.

At 120, the data required to represent the raw image is reduced. For example, the RGB information for each pixel in an image may be reduced so that instead of each pixel having intensity values for each of R, G, and B, each pixel has only one intensity corresponding to R, G or B. By reducing the number of intensity values from three to one, the data required to represent the image is reduced to ⅓ of the data needed for a full color RGB image. In this way, the amount of data needed to represent an image is reduced to ⅓ while maintaining color sensitivity needed for colored objects in the image. Pixels may be selected to be R, G, or B based on a pattern such as a Bayer pattern which is further detailed with respect to FIG. 3.

At 130, the reduced image is preprocessed. The preprocessing is minimized to increase processing speed and reduce computational complexity. For example, demosaicing and white balancing may be eliminated and basic pre-processing such as image cropping and resizing may be maintained.

At 140, perception is performed on the preprocessed image. Perception results include object bounding boxes. See, for example, FIG. 4 at 420. The performed perception is modified since the input images are not three channels for RGB, but are instead reduced to one channel of R, G, or B for each pixel. The perception is computerized and may occur at real time speeds in a moving vehicle without human assistance or feedback.

At 150, the perception result is provided as an output. The output may be used by further image processing tasks related to identifying objects and controlling the vehicle to avoid the objects.

Advantages of the disclosed techniques include the generation of a one-channel image compared to the three-channels (RGB) for usual images. This reduces the space required for storage by ⅔, and reduces the data rate or time for transmission, or a combination of data rate and time for transmission. The second advantage is reduced computational requirements because the image has less data to process and many preprocessing steps are eliminated. Another advantage is that the reduced image causes performance improvement because the raw data even though reduced has more information due to the reduced preprocessing (e.g., removed white balancing which may cause over-exposure).

FIG. 2 at 210 depicts an example of a pattern of colors assigned to pixels. Full color images contain three channels—red (R), green (G), and blue (B). FIG. 2 includes labels, “R” for red, “B” for blue, and “G” for green to clearly indicate the colors of the pixels and the colors of light indicated at 220. For every pixel in a typical image, there are three corresponding values. Together, they mix into a real ‘color point’ we see in the digital image. As described above, instead of three values per pixel, the number of values may be reduced to one by choosing R, G, or B for each pixel. The choice of which pixels are assigned to each may be selected in a pattern. For example, a Bayer pattern image has only one channel but encodes a subset of the information of the three RGB channels in the one channel. For example, a repeating pattern of 2*2 pixel grids may be selected. In some example embodiments, a ‘RGGB’ Bayer pattern may be used. RGGB refers to red for pixel 211, green for pixel 212, green for pixel 213, and blue for pixel 214 in a repeating 4-pixel square pattern as shown at 210. Different patterns may be selected depending on the hardware. In this way, the three values for pixel 211 corresponding to an R value, a G value, and a B value, the R value is selected and the G and B values are discarded. This same elimination process is applied to each pixel. The pattern may include a greater number of green pixel values than both red and blue. The pattern may be selected such that a value of every other pixel along a row in the reduced image corresponds to a green value of the raw image, or such that a value of every other pixel along a column in the reduced image corresponds to a green value of the raw image. The pattern may include a greater number of green pixel values than both red and blue. In a similar way as described above with respect to more frequent green pixels, red or blue pixels may be more frequent rather than green.

FIG. 2 at 220 shows the selection process from three values to one for an example pixel and the corresponding mosaic of each color selection on an image. The generating the reduced image may be performed using multiple color-selective filters. This may be referred to as a color filter array (CFA).

A Bayer pattern is an example of a color filter array (CFA). In some example embodiments, a color filter array different from a Bayer pattern can be used. Generally, the red, green, and blue colors used in a Bayer pattern can be transformed into another group of three colors where different combinations of the other group of three colors can be combined to cause the appearance of all other visible colors just as red, green, and blue can. Furthermore, a color filter array in some example embodiments may include four instead of three basic colors. For example, a patterned CYGM filter (cyan, yellow, green, magenta) can be used, or a patterned RGBE filter (red, green, blue, emerald) can be used as a CFA. Moreover, in some example embodiments, a CFA may add pixels that are not color filtered including a CMYW (cyan, magenta, yellow, and white) CFA.

FIG. 3 at 305 and 310 depict example images with an RGGB Bayer pattern applied to an original full color image. The pattern in the image at 310 is zoomed in to better show the pattern. FIG. 3 at 310 depicts a different image from the image at 305, 320 or in FIG. 4. The image at 305 with RGGB Bayer pattern is the same image as FIG. 3A with the R, G, or B value maintained as one color per pixel. FIG. 3A at 320 depicts a Bayer pattern image with the value for each R, G, and B in the pattern mapped to a black and white gray scale.

FIG. 4 at 410 depicts a rendering result after image processing is performed on the image shown at 320. FIG. 4 at 420 depicts a perception result showing object detection on the Bayer pattern image 305.

Some example implementations may be described as following examples.

1. A method for processing an image, comprising: receiving an image including an x-y array of pixels from a sensor array, each pixel in the x-y array of pixels having a value selected from one of three primary colors, based on a corresponding x-y value in a mask pattern; generating a preprocessed image by performing preprocessing on the image; and performing computerized perception on the preprocessed image to determine one or more outlines of physical objects.

2. The method of example 1, wherein the mask pattern is a Bayer pattern.

3. The method of example 2, wherein the Bayer pattern is a red-green-green-blue pattern assigned to a 2-pixel by 2-pixel array repeated across the raw image.

4. The method of example 1, wherein the pattern includes a greater number of green pixel values than both red and blue.

5. The method of example 1, wherein the pattern is selected such that a value of every other pixel along a row in the reduced image corresponds to green value of the raw image.

6. The method of example 1, wherein the pattern is selected such that a value of every other pixel along a column in the reduced image corresponds to green value of the raw image.

6. The method of example 1, wherein, each x-y value in the pattern is from one of three possible values.

7. The method of example 1, wherein the image is generated using one or more color-selective filters.

8. The method of example 1, wherein the preprocessing is performed on the image from the sensor array without human perception image enhancement.

9. The method of example 8, wherein the human perception image enhancement includes one or more of de-mosaicing, white balancing, and noise reduction.

10. The method of example 1, wherein the preprocessing does not include scaling one or more pixels' R, G, or B value for white balancing.

11. The method of example 1, wherein the preprocessing does not include reconstruction a full color image from incomplete color samples output from the sensor array overlaid with a color filter array for de-mosaicing.

12. The method of example 1, wherein the preprocessing does not include noise reduction, wherein noise reduction includes reduction of salt and pepper noise, wherein a noisy pixel bears little relation to the color of surrounding pixels, or reduction of Gaussian noise.

13. The method of example 1, wherein the preprocessing includes image cropping.

14. The method of example 1, wherein the preprocessing includes image resizing.

15. The method of example 1, wherein the preprocessing includes image compression.

16. The method of example 1, wherein the sensor array is a camera.

17. A computer apparatus comprising a processor, a memory and a communication interface, wherein the processor is programmed to implement a method recited in one or more of examples 1 to 16.

18. A computer readable program medium having code stored thereon, the code, when executed by a processor, causing the processor to implement a method recited in one or more of examples 1 to 16.

FIG. 5 depicts an example of an apparatus 500 that can be used to implement some of the techniques described in the present document. For example, the hardware platform 500 may implement the process 100 or may implement the various modules described herein. The hardware platform 500 may include a processor 502 that can execute code to implement a method. The hardware platform 500 may include a memory 504 that may be used to store processor-executable code and/or store data. The hardware platform 500 may further include a communication interface 506. For example, the communication interface 506 may implement one or more of the communication protocols (LTE, Wi-Fi, and so on) described herein.

Implementations of the subject matter and the functional operations described in this patent document can be implemented in various systems, digital electronic circuitry, or in computer software, firmware, or hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of one or more of them. Implementations of the subject matter described in this specification can be implemented as one or more computer program products, e.g., one or more modules of computer program instructions encoded on a tangible and non-transitory computer readable medium for execution by, or to control the operation of, data processing apparatus. The computer readable medium can be a machine-readable storage device, a machine-readable storage substrate, a memory device, a composition of matter effecting a machine-readable propagated signal, or a combination of one or more of them. The term “data processing unit” or “data processing apparatus” encompasses all apparatus, devices, and machines for processing data, including by way of example a programmable processor, a computer, or multiple processors or computers. The apparatus can include, in addition to hardware, code that creates an execution environment for the computer program in question, e.g., code that constitutes processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more of them.

A computer program (also known as a program, software, software application, script, or code) can be written in any form of programming language, including compiled or interpreted languages, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A computer program does not necessarily correspond to a file in a file system. A program can be stored in a portion of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub programs, or portions of code). A computer program can be deployed to be executed on one computer or on multiple computers that are located at one site or distributed across multiple sites and interconnected by a communication network.

The processes and logic flows described in this specification can be performed by one or more programmable processors executing one or more computer programs to perform functions by operating on input data and generating output. The processes and logic flows can also be performed by, and apparatus can also be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit).

Processors suitable for the execution of a computer program include, by way of example, both general and special purpose microprocessors, and any one or more processors of any kind of digital computer. Generally, a processor will receive instructions and data from a read only memory or a random-access memory or both. The essential elements of a computer are a processor for performing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto optical disks, or optical disks. However, a computer need not have such devices. Computer readable media suitable for storing computer program instructions and data include all forms of nonvolatile memory, media and memory devices, including by way of example semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.

While this patent document contains many specifics, these should not be construed as limitations on the scope of any invention or of what may be claimed, but rather as descriptions of features that may be specific to particular embodiments of particular inventions. Certain features that are described in this patent document in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination. Moreover, although features may be described above as acting in certain combinations and even initially claimed as such, one or more features from a claimed combination can in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.

Similarly, while operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. Moreover, the separation of various system components in the embodiments described in this patent document should not be understood as requiring such separation in all embodiments.

Only a few implementations and examples are described and other implementations, enhancements and variations can be made based on what is described and illustrated in this patent document.

Claims

1. A method for processing an image, comprising: receiving an image including an array of pixels from a sensor, each pixel in the array of pixels having a value selected from one of three primary colors based on a corresponding position value in a mask pattern;generating a preprocessed image by performing preprocessing on the image, wherein the preprocessing excludes a reduction of salt and pepper noise, or another reduction of Gaussian noise; andperforming perception on the preprocessed image.

2. The method of claim 1, wherein the mask pattern is a Bayer pattern.

3. The method of claim 2, wherein the Bayer pattern is a red-green-green-blue pattern assigned to a 2-pixel by 2-pixel array repeated across a raw image.

4. The method of claim 1, wherein the mask pattern includes a greater number of green pixel values than both red and blue.

5. The method of claim 1, wherein the mask pattern is selected such that a value of every other pixel along a row in a reduced image corresponds to green value of a raw image.

6. The method of claim 1, wherein the mask pattern is selected such that a value of every other pixel along a column in a reduced image corresponds to a green value of a raw image.

7. The method of claim 1, wherein, each position value in the mask pattern is from one of three possible values.

8. The method of claim 1, wherein the image is generated using one or more color-selective filters.

9. The method of claim 1, wherein the preprocessing is performed on the image from the sensor without human perception image enhancement.

10. The method of claim 9, wherein the human perception image enhancement includes one or more of de-mosaicing, white balancing, and noise reduction.

11. The method of claim 1, wherein the preprocessing excludes scaling for white balancing.

12. The method of claim 1, wherein the preprocessing excludes reconstruction of a full color image from incomplete color samples output from the sensor overlaid with a color filter for de-mosaicing.

13. An apparatus comprising: at least one processor and memory including executable instructions that when executed perform operations comprising:receiving an image including an array of pixels from a sensor, each pixel in the array of pixels having a value selected from one of three primary colors based on a corresponding position value in a mask pattern;generating a preprocessed image by performing preprocessing on the image, wherein the preprocessing excludes a reduction of salt and pepper noise, or another reduction of Gaussian noise; andperforming perception of the preprocessed image.

14. The apparatus of claim 13, wherein the mask pattern is a Bayer pattern.

15. The apparatus of claim 14, wherein the Bayer pattern is a red-green-green-blue pattern assigned to a 2-pixel by 2-pixel array repeated across a raw image.

16. The apparatus of claim 13, wherein the mask pattern includes a greater number of green pixel values than both red and blue.

17. The apparatus of claim 13, wherein the at least one processor selects the mask pattern such that a value of every other pixel along a row in a reduced image corresponds to a green value of a raw image.

18. A non-transitory computer readable medium having code stored thereon, the code, when executed by a processor, causing the processor to implement a method comprising: receiving an image including an array of pixels from a sensor, each pixel in the array of pixels having a value selected from one of three primary colors based on a corresponding position value in a mask pattern; andperforming perception on a preprocessed image that excludes a reduction of salt and pepper noise, or another reduction of Gaussian noise.

19. The non-transitory computer readable medium of claim 18, wherein preprocessing is performed on the image from the sensor without human perception image enhancement.

20. The method of claim 1, wherein the performing the perception on the preprocessed image determines one or more outlines of objects.