METHOD AND SYSTEM FOR HIGH-SPEED DUAL-VIEW BAND-LIMITED ILLUMINATION PROFILOMETRY

Abstract

A system and a method for 3D imaging of an object, the method comprising projecting sinusoidal fringe patterns onto the object using a projecting unit and capturing fringe patterns deformed by the object, alternatively by at least a first camera and a second camera, and recovering a 3D image of the object pixel by pixel from mutually incomplete images provided by the first camera and the second camera, by locating a point in images of the second camera that matches a selected pixel of the first camera; determining estimated 3D coordinates and wrapped phase based on calibration of the cameras, determining an horizontal coordinate on the plane of a projector of the projecting unit based on calibration of the projector, and using a wrapped phase value to recover a 3D point of 3D coordinates (x, y, z).

Description

FIELD OF THE INVENTION

The present invention relates to three-dimensional imaging. More specifically, it is concerned with a system and a method for high-speed dual-view band-limited illumination profilometry.

BACKGROUND OF THE INVENTION

Three-dimensional (3D) surface imaging has been extensively applied in a number of fields in industry, entertainment, and biomedicine. Among developed methods, structured-light profilometry has gained increasing popularity in measuring dynamic 3D objects due to high measurement accuracy and high imaging speeds. Phase-shifting fringe projection profilometry (PSFPP) for instance uses a set of sinusoidal fringe patterns as the basis for coordinate encoding, and, in contrast to other methods such as binary pattern projection for example, the pixel-level information carried by the phase of the fringe patterns is insensitive to variations in reflectivity across the surface of an the object's surface, which results in high accuracy in 3D measurements. The sinusoidal fringes are typically generated using digital micromirror devices (DMDs). Each micromirror on the digital micromirror devices can be independently tilted to either +12° or −12° from the normal to its surface to generate binary patterns at up to tens of kilohertz. Although they are binary amplitude spatial light modulators, it was shown that digital micromirror devices can be used to generate grayscale fringe patterns at high speeds. The average reflectance rate of each micromirror can be controlled by conventional dithering method to form a grayscale image. However, the projection rate of fringe patterns is limited at hundreds of hertz. To improve the projection speed, binary defocusing methods have been developed to produce a quasi-sinusoidal pattern by slightly defocusing a single binary digital micromirror device pattern. Nonetheless, the image is generated at a plane unconjugated to the digital micromirror device, which compromises the depth-sensing range and is less convenient to operate with fringe patterns of different frequencies. Recently, band-limited illumination was developed to control the system bandwidth by placing a pinhole low-pass filter at the Fourier plane of a 4f imaging system. Both the binary defocusing method and the band-limited illumination scheme allow generating one grayscale sinusoidal fringe pattern from a single binary digital micromirror device pattern. Thus, the fringe projection speed matches the refreshing rate of the digital micromirror device.

High-speed image acquisition is indispensable to digital micromirror device-based phase-shifting fringe projection profilometry. In standard phase-shifting fringe projection profilometry methods, extra calibration patterns must be used to avoid phase ambiguity, which reduces the overall 3D imaging speed. A solution to this problem is to use multiple cameras to simultaneously capture the full sequence of fringe patterns. The enriched observation of the 3D object eliminates the necessity of calibration patterns in data acquisition and phase unwrapping. This advancement, along with the incessantly increasing imaging speeds of cameras, has endowed multi-view phase-shifting fringe projection profilometry systems with image acquisition rates that keep up with the refreshing rates of digital micromirror devices.

Current multi-view phase-shifting fringe projection profilometry systems are still limited, mainly in two aspects. First, each camera must capture the full sequence of fringe patterns. This requirement imposes redundancy in data acquisition, which ultimately clamps the imaging speeds of systems. Given the finite readout rates of camera sensors, a sacrifice of the field of view (FOV) is inevitable for higher imaging speeds. Advanced signal processing approaches, such as image interpolation and compressed sensing, applied to mitigate this trade-off typically involve high computational complexity and reduced image quality. Second, generally the cameras are placed on different sides of the projector, and this arrangement may induce a large intensity difference from the directional scattering light and the shadow effect from the occlusion by local surface features, both of which reduce the reconstruction accuracy and exclude the application from non-Lambertian surfaces.

There is a need in the art for a method and a system for high-speed dual-view band-limited illumination profilometry.

SUMMARY OF THE INVENTION

More specifically, in accordance with the present invention, there is provided a system for 3D imaging of an object, the system comprising a projection unit and at least two projection unit cameras; the projection unit comprising a light source and a projector; the cameras being positioned on a same side of the projector; wherein the projection unit projects sinusoidal fringe patterns onto the object and the cameras alternatively capture, point by point; fringe patterns deformed by the object, depth information being encoded into the phase of the deformed fringe patterns, and the object being recovered by phase demodulation and reconstruction.

There is further provided a method for 3D imaging of an object, comprising projecting sinusoidal fringe patterns onto the object using a projecting unit and capturing fringe patterns deformed by the object, alternatively by at least a first camera and a second camera, and recovering a 3D image of the object pixel by pixel from mutually incomplete images provided by the first camera and the second camera, by locating a point in images of the second camera that matches a selected pixel of the first camera; determining estimated 3D coordinates and wrapped phase based on calibration of the cameras, determining an horizontal coordinate on the plane of a projector of the projecting unit based on calibration of the projector, and using a wrapped phase value to recover a 3D point of 3D coordinates (x, y, z).

Other objects, advantages and features of the present invention will become more apparent upon reading of the following non-restrictive description of specific embodiments thereof, given by way of example only with reference to the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

In the appended drawings:

FIG. 1A is a schematic view of a system according to an embodiment of an aspect of the present invention;

FIG. 1B is a timing diagram and acquisition sequence with exposure time t_e of the camera of the system of FIG. 1A;

FIG. 2 is a flow chart of a coordinate-based 3D point determination method with illustrative data according to an embodiment of an aspect of the present invention;

FIG. 3A shows 3D images of the planar surfaces (top images) and measured depth difference (bottom images) at different exposure times, the boxes representing the selected regions for analysis;

FIG. 3B shows depth resolution versus exposure time;

FIG. 4A shows reconstructed results of letter toys, with two perspective views (top row); selected depth profiles marked by dashed lines (bottom row) and close-up views;

FIG. 4B shows two perspective views of the reconstruction results of three toy cubes;

FIG. 5A shows reconstructed 3D images of a moving hand at five time points;

FIG. 5B shows movement traces of four fingertips, marked in the first panel in FIG. 5A;

FIG. 5C shows a front view of the reconstructed 3D image of a bouncing balls at different time points;

FIG. 5D shows the evolution of 3D positions of the bouncing balls, marked in the third panel in FIG. 5C;

FIG. 6A is a schematic view of an experimental setup; the field of view being marked by the dashed box;

FIG. 6B shows four reconstructed 3D images of the cup of FIG. 6A driven by a 500-Hz sound signal;

FIG. 6C shows evolution of the depth change of five points marked in of FIG. 6B with the fitted result;

FIG. 6D shows evolution of the averaged depth change with the fitted results under driving frequencies of 490 Hz, 500 Hz, and 510 Hz; error bar: standard deviation of Δz calculated from the five selected pixels;

FIG. 6E shows the response of the depth displacements to sound frequencies, the curve being a fitted result of a Lorentz function;

FIG. 7A shows six reconstructed 3D images showing a glass cup broken by a hammer;

FIG. 7B shows the evolution of 3D velocities of four selected fragments of the broken cup of FIG. 7A marked in the fourth and fifth panels in FIG. 7A; and

FIG. 7C shows the evolution of the corresponding 3D accelerations of the four selected fragments.

DESCRIPTION OF ILLUSTRATIVE EMBODIMENTS

The present invention is illustrated in further details by the following non-limiting examples.

A system for band-limited illumination profilometry (BLIP) with temporally interlaced acquisition (TIA) according to an embodiment of an aspect of the present invention generally comprises a projection unit to project pre-defined fringe patterns onto the surface of the measured object, the fringe patterns being distorted and reflected by the object surface, point by point, and cameras capturing the distorted fringes images, point by point.

In FIG. 1, the projecting unit comprises a high-coherent light source 10, a spatial light modulator 18, a pattern conversion unit 21, a projecting optics 30.

After expansion and collimation by a beam expander 12, the laser beam from a 10 Ins a 200-mW continuous-wave laser source 10, of wavelength λ=671 nm (MRL-III-671, CNI Lasers), is directed by mirrors 14 and 16 to a 0.45″ digital micromirror device 18 (AJD-4500, Ajile Light Industries) at an incident angle of about 24° to the normal of the surface of the digital micromirror device 18, for sinusoidal fringes generation; using four phase-shifting binary patterns, generated by an error diffusion algorithm from their corresponding grayscale sinusoidal patterns, loaded onto the digital micromirror device 18. The pattern conversion unit 21, comprising a 4f imaging system 25 with a low pass filter 24 such as a pinhole, converts the binary patterns to grayscale fringes at the intermediate image plane 28.

The minimal pinhole diameter D for all spatial frequency content of the sinusoidal fringe pattern to pass through the system is determined by the system bandwidth as follows:

$\begin{array}{cc}D=\frac{\lambda f_1}{p_f}& \left(1\right)\end{array}$

where p_f=324 μm is the period of the fringes composed by the digital micromirror device pixels, f₁ being the focal length of lens 22. With lenses 22 and 26 of the 4f imaging system 25 having focal lengths f₁=120 mm and f₂=175 mm respectively, the minimal pinhole diameter is D=248.52 μm. In an experiment, a 300 μm-diameter pinhole was selected.

A projector lens 30 (AF-P DX NIKKOR, Nikon) projects the output fringe patterns on a 3D object 32.

Deformed structured images are captured alternately by two high-speed CMOS cameras 34, 36 (CP70-1HS-M-1900, Optronis) placed side by side, i.e on a same side of the projector. Depending on their roles in image reconstruction, the cameras are referred to as the main camera 34 and the auxiliary camera 36 respectively, as will be described hereinbelow. Synchronized by the trigger signal of the digital micromirror device, each camera captures half of the sequence (FIG. 1B). The acquired images from each camera are transferred to a computer 38 via a CoaXPress cable connected to a frame grabber (Cyton-CXP, Bifflow).

The light source 10 is a high coherent light source of power at least 50 mW, selected depending on the sensitivity of cameras 34 and 36. with a laser wavelength is comprised in the range between about 380 and about 750 nm in case of visible light cameras, and in the range between about 800 and about 1100 nm in case of near infrared (NIR) cameras.

The high-speed cameras 34, 36 may be cameras with global shutter, of imaging speed of at least about 2 k frames/second, with image resolution at least about 1000×800 pixels.

The spatial light modulator 18 has a refreshing rate of at least about 4 kHz, on board memory of at least about 1 Mb, and is selected to work at the corresponding wavelength of the light source. It may be a liquid crystal display or a binary fringe mask with a motorized translation stage for example.

The pattern conversion unit 21 may comprise a 4f imaging system 25 with lenses of different focal lengths and the low-pass filter 24 may be a slit. The focal lengths of the two lenses are selected with a ratio (focal length of the first lens/focal length of the second length) comprised in the range between about 0.75 and about 1.5. The diameter of the low pass filter is selected in the range between about 150 μm and about 300 μm.

The projecting optics 30 is selected with a focal length in the range between about 18 and about 55 mm, a F number in the range between about 3.5 and about 5.6, and a magnification ratio in a range between about 5 and about 10 times.

The imaging speed and field of view may be further improved by using more than two cameras, in such a way to separate the workload to an array of cameras, for example to trace and recognize hand gesture in 3D space to provide information for human-computer interaction.

The system thus projects sinusoidal fringe patterns onto the object and captures the corresponding deformed patterns modulated by the object surfaces. The depth information is encoded into the phase of the distorted fringe images. For phase demodulation and reconstruction of the 3D object, the retrieved phase distribution corresponding to the object height is mathematically wrapped to principle values of arctangent function ranging between −π and π, and consequently, the phase discontinuities occur at the limits every time when the unknown true phase changes by 2π, which is referred to as the phase ambiguity problem, resulting from the periodical nature of the sinusoidal signal. A unique pixel correspondence between the cameras and the projector is obtained by phase unwrapping.

According to an aspect of the present disclosure, a method to recover the 3D image of the object pixel by pixel from the mutually incomplete images provided by the cameras generally comprises locating a point (u′_a, v′_a) in the images of the auxiliary camera 36 that matches a selected pixel (u_m, v_m) of the main camera 34; determining estimated 3D coordinates and wrapped phase from knowledge of the cameras calibration, determining the horizontal coordinate on the plane of the projector from knowledge of the projector calibration, and using the wrapped phase value to recover the 3D point of 3D coordinates (x, y, z) with the coordinate-based method.

System Calibration

To recover the object's 3D information, the method relies on a coordinate-based understanding of the spatial relationship between the projector 30 and the cameras 34, 36 in image formation. The projection of the 3D coordinates (x, y, z) of the 3D point onto the camera coordinates (u, v) is described in a pinhole model using extrinsic parameters R and T describing the rotation and translation of coordinates, respectively, and intrinsic parameters characterizing the properties of the cameras in image formation, with f_u and f_v the effective focal lengths along each axis of the sensor of the cameras; u_pp and v_pp the coordinates of the principal point of the cameras; and a accounting for pixel skewness, as follows:

$\begin{array}{cc}s\left[\begin{array}{c}u\\ v\\ 1\end{array}\right]=\left[\begin{array}{ccc}{f}_u& \alpha & u_{pp}\\ 0& f_v& v_{pp}\\ 0& 0& 1\end{array}\right]\left[\begin{array}{cc}R& T\end{array}\right]\left[\begin{array}{c}x\\ y\\ z\\ 1\end{array}\right],& \left(2\right)\end{array}$

Column vectors [u, v, 1]^T and [x, y, z, 1]^T represent the camera coordinates (u, v) and the 3D coordinates (x, y, z) in homogeneous coordinates, which allow for the numerical extraction of the camera coordinates (u, v) from Relation (2) through a scalar factor s.

The cameras and the projector are calibrated to determine the values of the extrinsic and intrinsic parameters using a checkerboard. Since the direct image acquisition is not possible for a projector, projector-centered images of the calibration object obtained by the phase-based mapping method are sent to a toolbox with calibration in the same manner as for the cameras.

Coordinate-Based 3D Point Determination

3D information is recovered from the calibrated imaging system using a coordinate-based method. To a point on the 3D object with the 3D coordinates (x, y, z) correspond two independent coordinates, (u, v) for the cameras and (u″, v″) for the projector.

In a calibrated phase-shifting fringe projection profilometry system, any three of these coordinates {u, v, u″, v″} can be determined and a linear system of the form E=M [x, y, z]^T is derived. The elements of E and M are obtained by using the calibration parameters of each device, the scalar factors and the three determined coordinates among u, v, u″ and v″. Thus, 3D information of an object point can be extracted via matrix inversion.

Returning to the system discussed in relation to FIG. 1A, first, images from the calibrated main camera 34 are used to provide the camera coordinates (u, v) of a point on the 3D object. Along with the calibration parameters of the system, an epipolar line is determined on the calibrated auxiliary camera 36. The horizontal coordinate in the images of the auxiliary camera 36 is recovered using search-based algorithms along the epipolar line, in stereo vision. Second, by substituting the calibrated projector in place of the auxiliary camera 36, the intensity values of the pixel (u, v) of the auxiliary camera 36 across a sequence of images is used by structured light methods to recover information about a coordinate of the calibrated projector. The object's 3D information is extracted pixel by pixel based on interlaced image acquisition by incorporating the camera coordinates (u,v) of a point on the 3D object and its corresponding projector coordinates, using a triangulation method to solve the point in 3D space.

Data Acquisition

For data acquisition, four fringe patterns with phases equally shifted by π/2 illuminate the 3D object. The intensity value I_k (u, v) for the pixel (u, v) in the kth image acquired by the calibrated main camera 34 is obtained as follows:

$\begin{array}{cc}{I}_k\left(u,v\right)=I_b\left(u,v\right)+I_{va}\left(u,v\right)\cos \left[\phi \left(u,v\right)-\frac{\pi k}{2}\right],& \left(3\right)\end{array}$

where k∈[0,3]. I_b(u, v) is the background intensity, I_va(u, v) is the variation of intensity and φ(u, v) is the depth-dependent phase.

Relation (3) allows analyzing two types of intensity matching conditions for the order of pattern projection shown in FIG. 1B. For a pixel (u′_a, v′_a) in the images of the auxiliary camera 36 that perfectly corresponds with the coordinates of a selected pixel (u_m, v_m) in the images of the main camera 34, Relation (3) yields:

I ₀(u_m,v_m)+I₂(u′_a,v′_a)=I₁(u_m,v_m)+I₃(u′_a,v′_a).  (4)

Rearrangement of Relation (4) leads to the equivalent relation, selected as the intensity matching condition:

I ₀(u_m,v_m)−I₁(u_m,v_m)=I₃(u′_a,v′_a)−I₂(u′_a,v′_a).  (5)

Each side of Relation (5) contains images captured by the same camera and represent a residual fringe component of sinusoidal characteristics, which allows to increase the efficiency of line-constrained searches by regularizing local maxima and minima in the patterns and by including additional phase information. Moreover, by considering the right-hand side as a continuously varying function along the epipolar line determined on the calibrated auxiliary camera 36, Relation (5) and bi-linear interpolation allows for the selection of discrete candidates with sub-pixel accuracy.

FIG. 2 is a flowchart of a method for coordinate-based 3D point determination according to an aspect of the present disclosure, with coordinates of the point to be matched for the main camera (u_m, v_m); coordinates of the estimated corresponding point for the auxiliary camera (u′_e, v′_e); recovered 3D coordinates (x, y, z); horizontal distance between the candidates and the estimated corresponding point r_i; phase value of the selected point in the main camera obtained by the Fourier transform profilometry method ω_m; phase value of the candidate points in the auxiliary camera obtained by the Fourier transform profilometry method; phase value obtained by the phase-shifting method φ′_ai; phase value determined on the projector's plane φ″_pi; 3D points determined by candidates P_i; principal point of the main camera P_m; principal point of the auxiliary camera P_a; ΔI_m=I₀ (u_m, v_m)−I₁(u_m, v_m); and intensity profile of I₃−I₂ along the epipolar line ΔI_ep.

In a quality map determination step (see Step I in FIG. 2), (I₀+I₁)/2 and (I₂+I₃)/2 are obtained from images by the main camera. Then, a threshold intensity, obtained from a selected background region, is used to eliminate pixels with low intensities and obtain a binary quality map. Subsequently, after such thresholding of the intensity map, only pixels (u_m, v_m) that fall within the quality map of the main camera are considered for 3D information recovery.

In a candidate discovery step (see Step II in FIG. 2), the selected pixel (u_m, v_m) of the main camera determines an epipolar line containing the matching point within the images of the auxiliary camera. Then, the candidates (u′_ai, v′_ai) for the matching point in the auxiliary camera images are extracted, the subscript “i” denoting the i^th candidate, that satisfy the intensity matching condition determined by Relation (5) above in addition to a quality map constraint, a transformation constraint and a phase sign constraint.

The quality map constraint requires that the candidates (u′_ai, v′_ai) for the matching point in the auxiliary images fall within the quality map of the auxiliary camera.

The transformation constraint requires that candidates occur within a segment of the epipolar line determined by a fixed two-dimensional projective transformation or homography that approximates the location of the matching point (u′_e, v′_e) within the images of the auxiliary camera as follows:

s′[u′_e,v′_e,1]^T=H[u_m,v_m,1]^T,  (6)

where s′ is a scalar factor representing extraction of the pair of coordinates of the estimated corresponding point (u′_e, v′_e) from its homogeneous coordinates [x, y, z, 1]^T. H is obtained by applying Relation (6) to four points chosen as the corners of a flat rectangular plane when imaged by both cameras at the approximate center of the measurement volume. [u_m, v_m, 1]^T are the homogeneous coordinates of the selected pixel (u_m, v_m) of the main camera. Once the coordinates of the estimated corresponding point (u′_e, v′_e) are determined, the search along the epipolar line is confined to the segment occurring over the horizontal interval [u′_e−r₀, u′_e+r₀], where r₀ is an experiment-dependent constant. In general, r₀ is selected as small as possible while still covering the targeted depth range. For the presently described experiments, the value of r₀ was set to 40 pixels.

The phase sign constraint requires that the selected point (u_m, v_m) of the main camera and candidates (u′_ai, v′_ai) have the same sign of their wrapped phases ω_m and ω′_ai respectively. Estimates of the wrapped phases are obtained using Fourier transform profilometry. In particular, the intensity I_f(u_m, v_m) of the selected pixel (u_m, v_m) of the main camera pixel in the filtered image is obtained by band-pass filtering the left-hand side of Relation (5) I₀−I₁, as follows:

$\begin{array}{cc}{I}_f\left(u_m,v_m\right)=\frac{\sqrt{2}}{2}{I}_{va}\left(u_m,v_m\right)\exp \left[j\left(\phi \left(u_m,v_m\right)+\frac{\pi }{4}\right)\right].& \left(7\right)\end{array}$

The wrapped phase estimation ω_m of the selected point (u_m, v_m) is obtained as follows:

$\begin{array}{cc}{\omega }_m={\tan }^{-1}\left\{\frac{\left[I_f\left(u_m,v_m\right)\right]}{\left[I_f\left(u_m,v_m\right)\right]}\right\},& \left(8\right)\end{array}$

where [⋅] and [⋅] denote the imaginary and real part of a complex variable respectively. The same band-pass filtering applied to the right-hand side of Relation (5) I₃−I₂ yields the estimate of its wrapped phase ω′_ai of the candidate (u′_ai, v′_ai), as follows:

$\begin{array}{cc}{\omega }_{ai}^{\prime }={\tan }^{-1}\left\{\frac{\left[I_f^{\prime }\left(u_{\mathrm{ai}}^{\prime },v_{ai}^{\prime }\right)\right]}{\left[I_f^{\prime }\left(u_{\mathrm{ai}}^{\prime },v_{ai}^{\prime }\right)\right]}\right\}.& \left(9\right)\end{array}$

The phase sign constraint requires that the wrapped phase estimation ω_m of the selected point (u_m, v_m) and the wrapped phase estimation ω′_ai of the candidate (u′_ai, v′_ai) have the same sign in the interval (−π, π].

Other Fourier transform profilometry methods for wrapped phase value extraction.

The output of the candidate discovery step is a pool of candidates for further evaluation and the method proceeds to matching point selection. If no candidate is found, the candidate discovery step is re-initiated for the next pixel in the main camera, until a candidate is obtained, and the method proceeds to the matching point selection.

In the matching point selection step (see Step III in FIG. 2), penalty scores for each candidate obtained from the candidate discovery step are determined. A first and primary criterion compares the phase values of the candidates using two methods. First, the phase of the candidate is obtained from the intensities of the candidate (u′_ai, v′_ai) and of the pixel (u_m, v_m) of the selected point as follows:

$\begin{array}{cc}{\phi }_{ai}^{\prime }={\tan }^{-1}\left[\frac{I_1\left(u_m,v_m\right)-I_3\left(u_{\mathrm{ai}}^{\prime },v_{ai}^{\prime }\right)}{I_0\left(u_m,v_m\right)-I_2\left(u_{\mathrm{ai}}^{\prime },v_{ai}^{\prime }\right)}\right].& \left(10\right)\end{array}$

Meanwhile, for each candidate (u′_ai, v′_ai), the coordinate triple (u_m, v_m, u′_ai) and knowledge of camera calibration allows determining an estimated 3D point P_i by using the stereo vision method. In addition, with the knowledge of the projector calibration, a point with coordinates (u″_pi, v″_pi) on the plane of the projector is determined for each candidate. Then, an unwrapped phase value φ″_pi is obtained by:

$\begin{array}{cc}{\phi }_{pi}^{″}=\frac{2\pi }{p}\left(u_{pi}^{″}-u_d^{″}\right),& \left(11\right)\end{array}$

where u″_d is a horizontal datum coordinate on the plane of the projector associated with the zero phase, and p is the fringe period in units of projector pixels. Since these independently obtained phase values must agree if the candidate correctly matches (u_m, v_m), a penalty score A_i, as a normalized difference of these two phase values, is obtained as follows:

$\begin{array}{cc}{A}_i=\frac{R\left({\phi }_{ai}^{\prime }-{\phi }_{pi}^{″}\right)}{\pi },& \left(12\right)\end{array}$

where the rewrapping function R(⋅) computes the subtracted difference between wrapped and unwrapped phase values.

To improve the robustness of the method, two additional criteria are implemented using data available from the candidate discovery step. B_i is a normalized distance score favoring candidates located closer to the estimated matching point (u′_e, v′_e), which is obtained by:

$\begin{array}{cc}{B}_i=\frac{u_e^{\prime }-u_{ai}^{\prime }}{r_0}.& \left(13\right)\end{array}$

Moreover, C_i is a normalized difference of wrapped phase values obtained by using the wrapped phases ω_m and ω′_ai, as follows:

$\begin{array}{cc}{C}_i=\frac{R\left({\omega }_m-{\omega }_{ai}^{\prime }\right)}{\pi }.& \left(14\right)\end{array}$

A total penalty score S_i for each candidate is then determined as a weighted linear combination of three individual scores as follows:

S _i=η₁A_i+η₂B_i+η₃C_i,  (15)

where the normalized weights [η₁, η₂, η₃]=[0.73, 0.09, 0.18] are empirically selected to lead to the results that are most consistent with physical reality. Finally, the candidate with the minimum total penalty score S_i is selected as the matching point (u′_a, v′_a), and its phase values are obtained by using relations. (10) and (11) are denoted as φ′_a and φ″_p, respectively.

In a final step of 3D point recovery (see Step IV in FIG. 2), the method determines the final 3D coordinates. First, the phase of the candidate φ′_a is unwrapped as φ′_a+2πq, where q is an integer such that φ″_p−(φ′_a+2πq)∈(−π,π]. Then, the coordinate on the plane of the projector u″_p is obtained with sub-pixel resolution as follows:

u″ _p =u″ _d +P(φ′_a/2π+q),  (16)

from which the final 3D coordinates (x, y, z) are obtained using calibration information associated with the coordinate triple (u_m, v_m, u″_p).

Results

FIG. 3 show quantification of the depth resolution of the method. To quantify the depth resolution with different exposure times, two stacked planar surfaces offset by about 9° were imaged. Reconstructed results at four representative exposure times t_e are shown in FIG. 3A. One area on each surface, marked as white boxes in full lines in FIG. 3A, was selected in the reconstructed image. The depth information on the x axis was obtained by averaging the depth values along the y axis. The difference in depths between these two surfaces is denoted by z_d. In addition, the noise is defined as the averaged values of the standard deviation in depth from both surfaces. The depth resolution is defined as when z_d equals to two times the noise level of the system. As shown in the four plots of FIG. 3A, the reconstruction results deteriorate with shorter exposure times, manifested by increased noise levels and more points incapable of retrieving 3D information. As a result, the depth resolution degrades from 0.06 mm at t_e=950 μs to 0.45 mm at t_e=150 μs (FIG. 3B). At exposure time t_e=100 μs, the method fails in 3D measurements. The region of unsuccessful reconstruction prevails across most of the planar surfaces. The noise dominates the obtained depth difference, which is attributed to the low signal-to-noise ratio in the captured images.

To examine the feasibility of the method, various static 3D objects were imaged. First, two sets of 3D distributed letter toys that composed the words of “LACI” and “INRS” were imaged. FIG. 4 show static 3D objects. Shown in FIG. 4A, the two perspective views of the reconstructed results reveal the 3D position of each letter toy. The detailed surface structures are illustrated by the selected depth profiles (white dashed lines in FIG. 4A). A proof-of-concept experiment was also conducted on three cube toys with fine structures, with a depth of about 4 mm, on the surfaces. As can be seen in FIG. 4B, the detailed structural information of these cube toys is recovered.

Imaging of Dynamic 3D Objects

To verify high-speed 3D surface profilometry, the method was used to image two dynamic scenes: a moving hand and three bouncing balls. The fringe patterns were projected at 4 kHz. The exposure times of both cameras were t_e=250 μs. Under these experimental conditions, a 3D imaging speed of 1 thousand frames per second (kips), a field of view (FOV) of 150 mm×130 mm, corresponding to 1180×860 pixels in captured images, and a depth resolution of 0.24 mm were achieved.

FIG. 5A shows the reconstructed 3D images of the moving hand at five time points from 0 ms to 60 ms with a time interval of 15 ms. The high-speed 3D imaging allowed tracking the movements of four fingertips. As shown in FIG. 5B, all the four fingers have apparent movement in both the x axis and the z axis but stay relatively stationary in the y axis, which agrees with the experimental condition.

In the second experiment, three white balls, each of which was marked by a different letter on its surface, bounced in an inclined transparent container. FIG. 5C shows five representative reconstructed images from 8 ms to 28 ms with a time interval of 5 ms. The changes of the letter “C” on B₁ and the letter “L” on B₂, marked in the third panel of FIG. 5C, clearly show the rotation of the two balls. The method enabled tracking the 3D centroids of each ball over time. As shown in FIG. 5D, B₁ collides with B₂ at 16 ms, resulting in a sudden change in the moving directions. This collision temporarily interrupted the free fall of B₁, represented by the two turning points in the curve of evolution along the y-axis (second panel of FIG. 5D). The collision also changed the moving direction of B₂, making it touch the base at 27 ms and then bounce up. In this scene, B₃ maintained its movement in a single direction in both the x axis and the z axis. It fell onto the based and bounced back at 16 ms, resulting in a turning point in its y-t curve. Because of the inclined bottom plane, the y-value of B₃ at 16 ms was smaller than that of B₂ at 27 ms.

Application to the Study of Sound-Induced Vibration on Glass

To highlight the broad utility of the method, sound-induced vibration on glass was imaged. In an experiment (FIG. 6A), a glass cup was fixed on a table. A function generator drove a speaker to produce single-frequency sound signals, from 450 Hz to 550 Hz with a step of 10 Hz through a sound channel placed close to the wall of the cup. To image the vibration dynamics, fringe patterns were projected at 4.8 kHz. The cameras had an exposure time of t_e=205 μs. This configuration enabled a 3D imaging speed of 1.2 kfps, a field of view (FOV) of 120 mm×110 mm, corresponding to 960×800 pixels in captured images, and a depth resolution of 0.31 mm. FIG. 6B shows four representative 3D images of the instantaneous shapes of the glass cup driven by the 500-Hz sound signal, showing the dynamic of structural deformation of the glass cup. The evolution of depth changes was analyzed using five selected points, marked by P_A to P_E in the first panel of FIG. 6B. A shown in FIG. 6C, the depth changes of the five points are in accordance, which is attributed to the rigidness of the glass.

Time histories of averaged depth displacements under different sound frequencies were further analyzed. FIG. 6D shows the results at the driving frequencies of 490 Hz, 500 Hz, and 510 Hz. Each result was fitted by a sinusoidal function with a frequency of 490.0 Hz, 499.4 Hz, and 508.6 Hz, respectively. These results show that the rigid glass cup vibrated in compliance with the driving frequency. Moreover, the amplitudes of fitted results, Δz_fit, were used to determine the relationship between the depth displacement and the sound frequency (FIG. 6E). This result was fitted by the Lorentz function, which determined the resonant frequency of this glass cup to 499.0 Hz.

Application to the Study of Glass Breakage

To further apply the method to recording non-repeatable 3D dynamics, the process of glass breaking by a hammer was imaged. As displayed in FIG. 7A, the growth of cracks and the burst of fragments with different shapes and sizes were clearly shown in the reconstructed 3D images. The time courses of velocities of four fragments, marked by F_A to F_D in FIG. 7A, are plotted in FIG. 7B. The velocities in the y axis are considerably small compared to the other two directions, which indicates the impact of the hammer force was exerted on the x-z plane. v_y of fragments F_A and F_C shows that they moved upward until 15 ms and fell afterward. v_y of fragments F_B and F_D reveals that they fell onto the remaining base of the cup at 15 ms and kept sliding down on the surface. The data of v_z illustrates that F_A and F_C moved closer to the cameras, which were directly driven by the hammer's force. However, F_B and F_D, which collided with other pieces, maintaining their positive directions in v_z to move away from the cameras. The corresponding accelerations are displayed as in FIG. 7C, which indicates the influence of both the main strike and the ensuing collision among different fragments. At 14 ms, the collision with other fragments, which applied an impact along the +x direction, dominated the acceleration direction for all four tracked fragments. In contrast, at 15 ms, another collision produced an impact in the −x direction, causing a sharp decrease in the acceleration for F_A and F_C. In addition, the direction of acceleration for F_D along the y-axis changed several times, which is attributed in several collisions of F_D with the base of the glass cup while sliding down.

There is thus presented a method with a kfps-level 3D imaging speed over a field of view of up to 150 mm×130 mm. The method implements temporally interlaced acquisition in multi-view 3D phase-shifting fringe projection profilometry systems, which allows each camera capturing half of the sequence of phase-shifting fringes. Leveraging the characteristics indicated in the intensity matching condition [Relation (5)], the method applies constraints in geometry and phase to find the matching pair of points in the main and auxiliary cameras and guides phase unwrapping to extract the depth information. The method was shown to allow the 3D visualization of glass vibration induced by sound and the glass breakage by a hammer.

There is thus presented a system and a method for high-speed dual-view band-limited illumination profilometry using temporally interlaced acquisition As people in the art will now be in a position to appreciate, temporally interlaced acquisition eliminates the redundant capture of fringe patterns in data acquisition. The roles of the main camera and the auxiliary camera are interchangeable and the present method may be adapted to a range of multi-view phase-shifting fringe projection profilometry systems. Moreover, temporally interlaced acquisition reduces the workload for both cameras by half. For the given bandwidth of the camera's interface, this more efficient use of cameras can either increase the 3D imaging speed for a fixed field of view or enlarge the field of view with a maintained 3D imaging speed. Both advantages shed light on implementing the present method with an array of cameras to simultaneously accomplishing high accuracy and high speed 3D imaging over a larger files of view. Also, the two cameras deployed in the present method are placed on a same side relative to the projector, which circumvents the intensity difference induced by the directional scattering light from the 3D object and reduces shadow effect by occlusion occurring when placing the cameras on different sides of the projector. As a result, robust pixel matching in the image reconstruction algorithm allows to recover 3D information on non-Lambertian surfaces.

The imaging speed and field of view may be optimized by separating the workload to four cameras, by using a faster digital micromirror device, and by using a more powerful laser. The image reconstruction toward real-time operation may be increased by further adapting the 3D point recovery method to four cameras and by using parallel computing to accelerate the calculation.

The present method may be integrated in structured illumination microscopy and frequency-resolved multi-dimensional imaging. The present method may also be implemented in the study of the dynamic characterization of glass in its interaction with the external forces in non-repeatable safety test analysis. As another example, the present method may be used to trace and recognize the hand gesture in 3D space to provide information for human-computer interaction. Furthermore, in robotics, the present method may provide a dual-view 3D vision for object tracking and reaction guidance. Finally, the present method can be used as an imaging accelerometer for vibration monitoring in rotating machinery and for behavior quantification in biological science.

Temporally interlaced acquisition thus integrated in a dual-view phase-shifting fringe projection profilometry system allows each camera capturing half of the sequence of phase-shifting fringes. Leveraging the characteristics indicated in the intensity matching condition, the method applies constraints in geometry and phase to find the matching pair of points in the main and auxiliary cameras and guides phase unwrapping to extract the depth information.

The present method and system eliminate the redundant capture of fringe patterns in data acquisition, which lifts the long-standing limitation in imaging speed for multi-view phase-shifting fringe projection profilometry, and allows reducing the workload of cameras, which enables the enhancement of either the 3D imaging speed or the imaging field of view. Dynamic 3D imaging of over 1 thousand frames per second on a field of view of up to 150×130 mm², corresponding to 1180×860 pixels in captured images, was demonstrated. Moreover, by putting the two cameras side by side on a same did of the projector, the present method and system circumvent the influence of directional scattering light and occlusion effect for more robust reconstruction, thereby expanding the application range of multi-view phase-shifting fringe projection profilometry to non-Lambertian surfaces.

The present method and system may be adapted into other multi-view 3D profilometers, thus opening new opportunities to blur-free 3D optical inspection and characterization with high speeds, large fields of view, and high accuracy. The present method and system provide a versatile tool for dynamic 3D metrology with potential applications in advanced manufacturing, such as characterization of glass in non-repeatable safety test and high-speed vibration monitoring in rotating machinery. The present compact and symmetric system may be embedded in the vision system of robots to track objects, to recognize the gesture for human-computer interaction, and to guide reactions.

The scope of the claims should not be limited by the embodiments set forth in the examples, but should be given the broadest interpretation consistent with the description as a whole.

Claims

1. A system for 3D imaging of an object, the system comprising: a projection unit; said projection unit comprising a light source and a projector; andat least two projection unit cameras, said cameras being positioned on a same side of said projector;wherein said projection unit projects sinusoidal fringe patterns onto the object and the cameras alternatively capture, point by point; fringe patterns deformed by the object, depth information being encoded into the phase of the deformed fringe patterns, and the object being recovered by phase demodulation and reconstruction.

2. The system of claim 1, wherein said light source is a high-coherent light source of a power of at least 50 mW.

3. The system of claim 1, wherein said cameras have an imaging speed of at least 2 k frames/second, and image resolution at least 1000×800 pixels.

4. The system of claim 1, wherein said projection unit comprises a spatial light modulator, and said spatial light modulator has a refreshing rate of at least 4 kHz and on board memory of at least 1 MB.

5. The system of claim 1, wherein said projection unit comprises a spatial light modulator, and said spatial light modulator is one of: a digital micromirror device, a liquid crystal display and a binary fringe mask with a motorized translation stage.

6. The system of claim 1, wherein said projection unit comprises a pattern conversion unit, and said pattern conversion unit comprises a 4f imaging system and a low-pass filter.

7. The system of claim 1, wherein said projection unit comprises a pattern conversion unit, and said pattern conversion unit comprises a 4f imaging system and a low-pass filter, a ratio between a focal length of a first lens and a focal length of a second lens of the 4f imaging system being selected in a range between 0.75 and 1.5, and a diameter of the low pass filter being selected in a range between 150 μm and 300 μm.

8. The system of claim 1, wherein said projection unit comprises a pattern conversion unit, and said pattern conversion unit comprises a 4f imaging system and a low-pass filter, said low pass filter being one of: a pinhole and a slit.

9. The system of claim 1, wherein said projection unit comprises a pattern conversion unit, and said pattern conversion unit comprises a 4f imaging system and a low-pass filter, said low pass filter being one a pinhole, said pinhole having a minimal diameter determined by: D=λf1/pf, where p is a period of the fringes and f1 is a focal length of a first lens of the 4f imaging system.

10. The system of claim 1, wherein said projector is selected with a focal length in a range between 18 and 55 mm, a F number in a range between 3.5 and 5.6, and a magnification ratio in a range between 5 and 10 times.

11. A method for 3D imaging of an object, comprising: projecting sinusoidal fringe patterns onto the object using a projecting unit and capturing fringe patterns deformed by the object, alternatively by at least a first camera and a second camera, andrecovering a 3D image of the object pixel by pixel from mutually incomplete images provided by the first camera and the second camera, by locating a point in images of the second camera that matches a selected pixel of the first camera; determining estimated 3D coordinates and wrapped phase based on calibration of the cameras, determining an horizontal coordinate on the plane of a projector of the projecting unit based on calibration of the projector, and using a wrapped phase value to recover a 3D point of 3D coordinates (x, y, z).

12. The method of claim 11, comprising calibrating the first and second cameras and the projecting optics, and recovering 3D information using a coordinate-based method, wherein, to the point on the object with the 3D coordinates (x, y, z) correspond two independent coordinates, (u, v) for the cameras and (u″, v″) for the projector.

13. The method of claim 11, comprising calibrating the first and second cameras and the projecting optics, using images captured by the first camera to provide camera coordinates of a point on the object, and determining an epipolar line on the second camera, recovering an horizontal coordinate in images captured by the second camera along the epipolar line; by substituting the projector in place of the second camera, using intensity values of the camera coordinates of the point of the second camera across a sequence of images to recover information about a coordinate of the projector; and extracting 3D information of the object pixel by pixel based on interlaced image acquisition by incorporating the camera coordinates (u,v) of the point on the 3D object and its corresponding projector coordinates.

14. The method of claim 11, comprising: obtaining a threshold intensity from a selected background region in images captured by the first camera, using the threshold intensity to eliminate pixels in the images captured by the first camera to obtain a binary quality map of the first camera, and selecting a pixel of an estimated corresponding pixel in images captured by the second camera within the quality map of the first camera;the selected pixel of the first camera determining an epipolar line containing a matching point within the images of the second camera, extracting candidates for the matching point in the second camera mages that satisfy an intensity matching condition.

15. The method of claim 11, comprising: obtaining a threshold intensity from a selected background region in images captured by the first camera, using the threshold intensity to eliminate pixels in the images captured by the first camera to obtain a binary quality map of the first camera, and selecting a pixel in images of the second camera within the quality map of the first camera as a candidate;the selected pixel of the first camera determining an epipolar line containing the matching point within the images of the second camera, extracting candidates for the matching point in the second camera images that satisfy an intensity matching condition;determining penalty scores for each obtained candidate; anddetermining final 3D coordinates, by unwrapping the phase of the candidates and obtaining a coordinate on the plane of the projector.

16. The method of claim 11, comprising: obtaining a threshold intensity from a selected background region in images captured by the first camera, using the threshold intensity to eliminate pixels in the images captured by the first camera to obtain a binary quality map of the first camera, and selecting a pixel in images of the second camera within the quality map of the first camera as a candidate;the selected pixel of the first camera determining an epipolar line containing the matching point within the images of the second camera, extracting candidates for the matching point in the second camera images that satisfy an intensity matching condition and quality map constraint, transformation constraint and phase sign constraint; the quality map constraint requiring that the candidates for the matching point in the second camera images fall within the quality map of the second camera; the transformation constraint requiring that the candidates occur within a segment of the epipolar line; and the phase sign constraint requiring that the selected point of the first camera and the candidates have a same sign of respective wrapped phases;determining penalty scores for each obtained candidate; anddetermining final 3D coordinates, by unwrapping the phase of the candidates and obtaining a coordinate on the plane of the projector.

17. The method of claim 11, comprising positioning the cameras on a same side of the projector.

18. The method of claim 11, wherein the projection unit comprises a light source, a spatial light modulator, and a pattern conversion unit, the spatial light modulator generating sinusoidal fringes from beams from the light source using pre-defined binary patterns, and the pattern conversion unit converting the binary patterns to grayscale fringes; the cameras capturing alternatively deformed structured images; depth information being encoded into the phase of the deformed images, and the object being recovered by phase demodulation and reconstruction.

19. The method of claim 11, wherein the projection unit comprises a light source, a spatial light modulator and a pattern conversion unit, the spatial light modulator generating sinusoidal fringes from beams from the light source using pre-defined binary patterns, and the pattern conversion unit converting the binary patterns to grayscale fringes; the cameras capturing alternatively deformed structured images; depth information being encoded into the phase of the deformed images, and the object being recovered by phase demodulation and reconstruction, the light source being a high-coherent light source of a power of at least 50 mW, the cameras having an imaging speed of at least 2 k frames/second, and image resolution at least 1000×800 pixels, the spatial light modulator having a refreshing rate of at least 4 kHz and on board memory of at least 1 Mb, and the projector being selected with a focal length in a range between 18 and 55 mm, a F number in a range between 3.5 and 5.6, and a magnification ratio in a range between 5 and 10 times.

20. The method of claim 11, wherein the projection unit comprises a light source, a spatial light modulator, and a pattern conversion unit, the spatial light modulator generating sinusoidal fringes from beams from the light source using pre-defined binary patterns, and the pattern conversion unit converting the binary patterns to grayscale fringes; the cameras capturing alternatively deformed structured images; depth information being encoded into the phase of the deformed images, and the object being recovered by phase demodulation and reconstruction, the light modulator being one of: a digital micromirror device, a liquid crystal display and a binary fringe mask with a motorized translation stage, and the pattern conversion unit comprising a 4f imaging system and a low-pass filter, with a ratio between a focal length of a first lens and a focal length of a second lens of the 4f imaging system comprised in a range between 0.75 and 1.5, and a diameter of the low pass filter being selected in a range between 150 μm and 300 μm.