Non-vision field imaging simulation system based on filtered back projection and reconstruction method

Abstract

The invention belongs to the field of non-vision field imaging, and particularly relates to a non-vision field imaging system based on filtered back projection and a reconstruction method. Comprising the following steps: 1) setting scene parameters based on a non-vision field imaging requirement; the imaging scene in the step 2) is mainly composed of pulse laser, a shielding wall surface, a relay wall surface, objects outside the sight line, a high-time-resolution device and the like; (3) the pulse laser is focused on the relay wall surface, the pulse laser is emitted, and dot matrix scanning is carried out; (4) the relay wall carries out diffuse reflection on the pulse laser, and the pulse laser irradiates the surface of the hidden object; 5) an imaging module collects diffuse reflection information to obtain an intensity-time 2D space image; and 6) the data processing module carries out filtering back projection calculation processing to obtain hidden object surface information corresponding to each voxel, and outputs the information as a reconstructed image. The method is low in cost, low in hardware equipment requirement and convenient to implement, and the generated reconstructed image is visual.

Description

Technical Field

[0001] The present invention belongs to the field of non-line-of-sight imaging, and in particular relates to a non-line-of-sight imaging system and a reconstruction method based on filtered back projection. Background Art

[0002] As an emerging technology that has attracted much attention, non-line-of-sight imaging technology can reconstruct hidden targets by analyzing the diffuse reflection of the relay surface, and has broad application prospects in the fields of autonomous driving and medical imaging.

[0003] Despite the challenges of low signal-to-noise ratio and high morbidity, non-line-of-sight imaging has developed rapidly in recent years. Most of the current non-line-of-sight imaging technologies use traditional physical models to establish imaging models through active or passive illumination and use reconstruction algorithms to restore hidden scenes. In addition, deep learning algorithms for non-line-of-sight imaging have also received widespread attention in recent years.

[0004] Within the line of sight, modern computer imaging technology can achieve high-quality imaging at relatively long distances. However, due to the inherent physical constraints of visible light, it is difficult for traditional optical imaging to see objects outside the line of sight. To break this limitation, non-line-of-sight (NLOS) imaging analyzes the diffuse reflection of relay walls to image hidden objects, which has a wide range of applications in many fields, such as medical imaging, autonomous driving, and robotic vision.

[0005] The method based on the time of flight of light can be used in non-line-of-sight imaging technology. In traditional optical imaging technology, the intensity of multiple diffuse reflections is low compared to the integrated intensity collected by the camera within a certain period of time, and is mixed with the signal in the field of view, so it is regarded as noise and rejected from the imaging system, thus not being fully utilized. Non-line-of-sight imaging uses such diffuse reflections to reconstruct the surface shape of the object.

[0006] Based on the photon flight time method, the weak and scattered light signals after multiple reflections are captured, and the arrival time of the photons is used to analyze the path of the light before it bounces multiple times to reach the detector, and the surface of the hidden object is reconstructed, thereby inferring the shape of the object hidden outside the line of sight, making imaging beyond the field of view possible. It is an important method in non-line-of-sight imaging technology.

[0007] High-quality time-of-flight cameras are now widely available, but programmable capture experiments remain extremely challenging, very expensive, and cumbersome. Commercial lidars are difficult to program and obtain temporal profiles because they are only suitable for recording the first bounce.

[0008] The present invention proposes an imaging simulation model of multiple bounces of light pulses in a non-viewing scene, and a method for restoring three-dimensional features from three-dimensional diffuse reflections by using the constructed model to infer the shape of objects outside the viewing area. Summary of the invention

[0009] A non-line-of-sight imaging simulation system and reconstruction method based on filtered back projection. The system and reconstruction method include: a non-line-of-sight imaging module and a data processing module; The simulation system includes the following main implementation steps: Step 1) Based on the requirements of the non-line-of-sight imaging scene, set the specific content of the non-line-of-sight imaging scene. The instrument parameters used, scene settings, and specific parameters of the data acquisition module simulation system; The imaging scene described in step 2) is generally composed of a pulsed laser, a galvanometer scanner, a focusing mirror, a beam splitter, a sight-blocking wall, a relay wall, an object hidden from sight, and a high time resolution device (such as a streak camera, a SPAD array, etc.); In the imaging scene of step 3), the pulse laser is focused on the relay wall surface, and the pulse laser is emitted to the relay wall surface in a grid shape with a fixed interval to perform dot scanning; Step 4) the relay wall diffusely reflects the received pulsed laser to generate a scattered light signal, and irradiates the surface of the hidden object outside the field of view; Step 5) Under multiple scans, the data collection module in the imaging module, i.e., the high time resolution device, respectively collects diffuse reflection light information obtained from each scan to obtain multiple intensity-time-2D spatial images; Step 6) The data processing module integrates the images collected by each scan into an intensity-time-3D spatial image containing one more dimension generated by the scan, i.e., data containing 5D information; Step 5) The data processing module performs filtered back-projection calculation processing on the 5D data to obtain the confidence value of each voxel as the surface information of the object outside the field of view corresponding to each voxel. This information can be visualized by the data processing module through a heat map and output as a reconstructed image.

[0010] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that the simulation system simulates a specific achievable non-line-of-sight imaging scene; By emitting pulsed lasers to laser scanning points on the relay diffuse reflection wall, signals for reconstructing imaging of target objects hidden outside the field of view are obtained from the sensing points of the relay diffuse reflection wall; According to the scanning detection method, the information about the propagated distance light is accurately determined based on the use of high-speed captured photons, and this information is used to infer hidden geometric structures; the method is based on the ideas of energy front propagation and tomography, constructs forward and inversion models of cubic diffuse scattering, and restores the three-dimensional shape of non-field of view objects based on tomography methods and sparsity-based methods; the algorithm of the method is based on Fresnel approximation, simplifies the problem, and can achieve inversion through the back propagation process; the method can be used to discover and reconstruct hidden geometric structures.

[0011] According to the non-line-of-sight imaging simulation system and reconstruction method according to claim 1, it is characterized in that the inversion and reconstruction imaging result of the non-line-of-sight three-dimensional target object is an intuitive shape map of the non-line-of-sight three-dimensional target reconstructed based on the filtered back projection algorithm according to the scanned collected data.

[0012] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that: 1) When considering the influence of any scene point on the recorded 5D light transmission, the laser Illuminate a point on the visible surface of a wall , generating a virtual unfocused pulse light source from each scene point The reflected light is unfocused by the virtual sensor The model uses the Huygens-Fresnel principle to model the propagation of a pulse wave from each scene point, with the beam intensity being proportional to the total plane cross-section of the wavefront. changes, producing hyperbolic fringes, or simply "fringes", which are captured by a streak camera or a single-photon avalanche camera; 2) For the light scattering in the second and third paragraphs of the model, the hidden object is modeled as Time to send pulse The receiver is modeled as an array of unfocused receivers that capture photons with high temporal resolution. , the modeling can be achieved by using a high temporal resolution camera focused on the wall pixels. To simplify further, the model assumes that the speed of light is constant and expresses other quantities in distance units. The following mathematical equations represent the data recorded by the high temporal resolution camera after making the above assumptions: in , , , , represents the time shift between the transmitter and the receiver, and represents the intensity observed at a certain sensor at a certain moment.

[0013] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that if the position of the transmitter is fixed (or the position of the laser is fixed) and the position of the receiver, the equation is an ellipsoid at the transmitter position, and if the observation is restricted to a plane, it is an ellipse. Under the above simplified conditions, the equation can be rewritten as: , this problem is similar to fault projection, and formula (3) can be rewritten as The world volume W(x) of the hidden object is a delta function with support on the reconstructed target surface, and the resulting single fringe image is the result of measuring the elliptical projection of the world volume W(x). This method is also called elliptical tomography by researchers.

[0014] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that when a streak camera is used as the main imaging hardware device: The streak camera has a high temporal resolution and can measure the temporal variation of incident light relative to position, i.e., perform time- and space-resolved measurements. The light pulse to be measured passes through a slit and forms a slit image on the photocathode via an optical element. In the illustrated case, the four light pulses vary slightly in time and space, and the light intensities of the four pulses are also different. After the light pulse reaches the photocathode, it is converted into a certain number of electrons, and the number of electrons is proportional to the intensity of the light pulse. The four light pulses shown in the figure are converted into electrons in turn, and then pass through a pair of accelerating electrodes (accelerating electrodes), which accelerate the electrons. The accelerated electrons bombard the phosphor screen. When the electrons pass through the paired scanning electrodes (sweep electrodes), a high voltage synchronized with the incident light time is applied to the scanning electrodes, and the electrons are scanned from top to bottom at high speed (sweep). The high-speed scanning makes the time when the electrons arrive at the phosphor screen slightly different, and they will produce different angles of deflection in the vertical direction, and then enter the micro-channel plate (micro-channel plate). When the electrons pass through the micro-channel plate, they are magnified thousands of times, and then bombard the phosphor screen, which converts the electrons into light signals again. The earliest light pulse is located at the top of the corresponding image, and the latest light pulse corresponds to the bottom image. The vertical direction of the phosphor screen is used as the time axis, and the horizontal position of each light pulse corresponds to the horizontal position at the time of incidence, and the brightness of the phosphor image is proportional to the intensity of each incident light pulse. The streak camera converts the time, space and intensity of the measured light into an image displayed on the fluorescent screen, thereby obtaining the corresponding time, space and intensity information of the incident light from the streak image.

[0015] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that when a SPAD array is used as the main hardware device for imaging: In the imaging system and method, the SPAD array has a linear change in leakage current as photons are absorbed, and carriers are released through the internal photoelectric effect. The photogenerated carriers of the SPAD are accelerated by the electric field in the device, generating kinetic energy sufficient to overcome the ionization energy of the material, impacting electrons to separate them from atoms, generating a large amount of avalanche current, and the current grows exponentially. Due to the high speed of avalanche establishment and the low timing jitter of the device, the SPAD can be used to obtain the arrival time of the incident photon. The SPAD is reverse biased at an operating voltage exceeding the junction breakdown voltage, and a single charge carrier injected into the depletion layer triggers a self-sustaining avalanche, and the current rises rapidly to a macroscopic stable level, where the output pulse indicates the trigger event. In Geiger mode, a single photon can trigger an avalanche of carriers, which can be used as a single photon counter and obtain fairly accurate photon timing. The temporal resolution of SPAD is lower than that of streak camera, but compared with streak camera, SPAD has a cost closer to consumer level. The impact ionization and avalanche levels of SPAD also make it more suitable for non-line-of-sight imaging scenarios with weak effective signals. In the hardware combination using pulsed laser and SPAD, scanning a series of points along the relay wall will produce a streak image. Each column in the image represents a histogram of photon arrival times measured at different discrete points on the relay wall, encoding spatiotemporal geometric information from hidden indirect light transmission.

[0016] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that when a SPAD array is used as the main hardware device for imaging: In the imaging system and method, the SPAD array has a linear change in leakage current as photons are absorbed, and carriers are released through the internal photoelectric effect. The photogenerated carriers of the SPAD are accelerated by the electric field in the device, generating kinetic energy sufficient to overcome the ionization energy of the material, impacting electrons to separate them from atoms, generating a large amount of avalanche current, and the current grows exponentially. Due to the high speed of avalanche establishment and the low timing jitter of the device, the SPAD can be used to obtain the arrival time of the incident photon. The SPAD is reverse biased at an operating voltage exceeding the junction breakdown voltage, and a single charge carrier injected into the depletion layer triggers a self-sustaining avalanche, and the current rises rapidly to a macroscopic stable level, where the output pulse indicates the trigger event. In Geiger mode, a single photon can trigger an avalanche of carriers, which can be used as a single photon counter and obtain fairly accurate photon timing. The temporal resolution of SPAD is lower than that of streak camera, but compared with streak camera, SPAD has a cost closer to consumer level. The impact ionization and avalanche levels of SPAD also make it more suitable for non-line-of-sight imaging scenarios with weak effective signals. In the hardware combination using pulsed laser and SPAD, scanning a series of points along the relay wall will produce a streak image. Each column in the image represents a histogram of photon arrival times measured at different discrete points on the relay wall, encoding spatiotemporal geometric information from hidden indirect light transmission.

[0017] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that when using the scene simulation model: Using density estimation techniques, the sampling paths are used to reconstruct the time-domain radiance, taking into account the distribution of radiance along time in the participating media. One of the most common assumptions in computer graphics is that the speed of light is considered to be infinite, which leads to the simulation of light transport in steady state. This is a reasonable assumption because most existing imaging hardware is very slow compared to the speed of light; The simulation model introduces a framework for transient imaging simulation, breaking the assumption that the speed of light is infinite. Using density estimation technology, the temporal changes of the signal can be reconstructed from the simulation samples. The simulation principle is based on a transient rendering theory framework based on path integrals, including propagation in free space as well as scattering on surfaces and in media. This enables the model to analyze transient rendering.

[0018] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that the participating medium sampling technology of the simulation model complements the traditional radiation sampling method with uniform sampling in the time domain. It provides time-resolved simulation of several light transport phenomena that cannot be observed in steady state. Transient rendering. The simulation model generalizes the rendering equation to a recursive operator, including the finite speed propagation of light.

[0019] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 are characterized in that the time domain is introduced into the photon mapping framework, which is achieved by adding a time smoothing kernel KT to the radiation estimation. Using M photons with contributions γi to estimate the radiation, the simulation system handles complex light paths in transient states, while gradually reducing the bias and variance introduced by progressive photon mapping and temporal density estimation, in the space and time domains, respectively. The classical path space integral is extended to include the time domain, and it is shown that the high-frequency characteristics of transient light transmission lead to serious sampling problems. The novel sampling strategy and density estimation technique enable it to evenly distribute samples in time, thereby reducing variance and maintaining a constant distribution of deviations.

[0020] According to the non-line-of-sight imaging simulation system and reconstruction method of claim 1, it is characterized in that in the reconstruction of tomographic images in actual applications, a stable and discrete version of the inverse Radon transform is usually used, called a filtered back projection algorithm. If only back projection is used, the image will be blurry. The process of filtered back projection can be understood as a combination of filtering and back projection processes, because the one-dimensional projection obtained in filtered back projection needs to be filtered by a one-dimensional Radon kernel to obtain a two-dimensional signal. The filter used does not contain a DC gain, so a DC bias may need to be added. Reconstruction using back projection will obtain better resolution than the interpolation method in the Fourier domain reconstruction algorithm. However, filtered back projection will cause greater noise because the one-dimensional Radon kernel filtering process tends to amplify high-frequency content. In the optical domain, filters can be described by wavelength rather than frequency. High pass and low pass have different meanings when characterized by wavelength than in circuits. "High pass" only passes longer wavelengths, that is, lower frequencies, while "low pass" only passes shorter wavelengths, that is, higher frequencies.

[0021] According to the non-line-of-sight imaging simulation system and reconstruction method of claim 1, it is characterized in that the first step of the imaging and reconstruction process is data acquisition, which is completed by scanning the grid on the relay wall. Each scanning process includes illuminating a point in the grid with a laser pulse, and recording the stripe image by the detector after the scattered pulse is reflected and enters the stripe camera after a very short time interval. It should be noted that the scanning process is time-sequential and multiple images may need to be collected to obtain a corrected image with less noise. The loaded stripe image then needs to be reprocessed to eliminate factors that affect the obtained data, such as time drift. Once the data is properly processed, the reconstruction algorithm can be applied to the stripe image. The reprocessing process includes time correction, intensity correction and non-uniform gain to remove ambient light, and in a non-confocal experimental setting, part of the laser can be directed onto the relay wall to complete the time correction. Filtering is applied after back-projection in the reconstruction stage, and threshold processing and compressed reconstruction are also involved, because the sampled data may be too dense to put a lot of pressure on the computer's processor, or even cause a crash, when the resulting data matrix is ​​sparsely populated. Finally, depending on the situation, the three-dimensional point cloud can be rendered in the threshold processing stage.

[0022] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that the imaging and reconstruction process includes three stages: 1) Stage 1: Data acquisition. The laser is pointed at different positions of the relay wall and the corresponding stripe images are captured. For each image in different positions, multiple images are taken and overlapped to reduce noise. 2) Stage 2: Data preprocessing. The stripe image is loaded and intensity corrected and shifted to adjust for spatiotemporal jitter. 3) Stage 3: 3D reconstruction. The clean stripe image is used to reconstruct the unknown shape.

[0023] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that, for each voxel and each fringe image in the working volume, the fringe image pixels that the voxel under consideration may contribute are calculated. Specifically, the voxel at position v may contribute to the pixel corresponding to point w on the wall at time t if Where C is the projection center of the camera and L is the laser position as described above. Stripe image pixels that satisfy this condition are called contributing pixels. Functions and heat maps are calculated on the voxel space.

[0024] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that the heat map is a function on a three-dimensional voxel grid. Assuming that the axes of the grid are ordered, the filtered heat map is calculated as the derivative of the heat map of the third axis of the voxel grid. The filtered heat map measures and calculates the confidence of the surface patch that there is a hidden object at a specific voxel location.

[0025] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that the sliding window maximum of the filtered heat map is calculated. The estimate of the 3D shape to be reconstructed consists of those voxels that meet the conditions.

[0026] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that the compressed reconstruction uses techniques such as SPGL1 and CoSAMP as an alternative to backprojection and filtering. It relies on the fact that the 3D voxel grid is sparsely populated and then rendered. The result of the thresholding step is a 3D point cloud. This point cloud can be visualized using Chimera rendering software.

[0027] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that another method that can be used for reconstruction is to model the captured data as a collection of hyperbolic fringes with unknown parameters. This problem can be formulated as an optimization problem, which can be achieved by imposing sparse constraints on the reconstructed world.

[0028] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that I(x, t) is a recorded fringe image, x and t are parameters of the fringe corresponding to valid pixels, and I is the corresponding intensity. Sparsity-based reconstruction uses SPGL1 and other compressed sensing techniques to reconstruct the hidden volume. In order to reconstruct the 2D manifold of the hidden object, the solution can be regarded as a sparse set of points within the voxelized world volume.

[0029] The non-field of view imaging simulation system and reconstruction method according to claim 1 is characterized in that the linearization system considers an isolated voxel in the world and focuses on the response it produces in the stripe image. The response changes orientation in time and space based on the laser position and the observed wall pixel. The response observed at each pixel on the wall is the linear time-invariant sum of the time signals sent by all active world voxels. A large matrix can be constructed to represent this system, with matrix A representing the corresponding linear transformation. Each column of the matrix represents the stripe image from the corresponding world voxel. Each row in the corresponding linear transformation matrix A identifies a unique pixel, laser-position pair. The matrix can be used to quickly draw stripe images and back-project. Since the number of wall pixels and laser positions is much larger than the number of world voxels, the size of the corresponding linear transformation matrix A can be reduced by subsampling the world voxels and selecting only a few lasers.

[0030] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that the algorithm CoSAMP can be used to solve the problem in the presence of noise. A very similar formulation is called basis pursuit denoising and involves minimizing a norm with constraints on the observed noise.

[0031] The non-line-of-sight imaging simulation system and reconstruction method according to claim 1 is characterized in that higher output power usually increases the intensity of the light illuminating the scene, which may cause safety problems and damage the scene material in extreme cases. The intensity can be reduced by increasing the diameter of the laser beam. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Set up examples for four typical NLOS imaging scenarios.

[0032] Figure 2 A schematic diagram of a hidden three-dimensional rabbit-shaped object in a non-viewing field imaging scene of the present invention (left), and a reconstructed image of the data output adopted by the present invention (right).

[0033] Figure 3 A schematic diagram of a hidden three-dimensional toy model object in a non-viewing field imaging scene of the present invention (left), and a reconstructed image output from the data collected by the present invention (right).

Claims

1. A non-line-of-sight imaging simulation system and reconstruction method based on filtered back projection. The system and reconstruction method include: Non-line-of-sight imaging module and data processing module; The simulation system includes the following main implementation steps: Step 1) Based on the requirements of the non-line-of-sight imaging scene, set the specific content of the non-line-of-sight imaging scene. The instrument parameters used, scene settings, and specific parameters of the data acquisition module simulation system; The imaging scene described in step 2) is generally composed of a pulsed laser, a galvanometer scanner, a focusing mirror, a beam splitter, a sight-blocking wall, a relay wall, an object hidden from sight, and a high-time-resolution device (such as a streak camera, a SPAD array, etc.). FIG1 shows a general non-viewing imaging scene setting; In the imaging scene of step 3), the pulse laser is focused on the relay wall surface, and the pulse laser is emitted to the relay wall surface in a grid shape with a fixed interval to perform dot scanning; Step 4) the relay wall diffusely reflects the received pulsed laser to generate a scattered light signal, and irradiates the surface of the hidden object outside the field of view; Step 5) Under multiple scans, the data collection module in the imaging module, i.e., the high time resolution device, respectively collects diffuse reflection light information obtained from each scan to obtain multiple intensity-time-2D spatial images; Step 6) The data processing module integrates the images collected by each scan into an intensity-time-3D spatial image containing one more dimension generated by the scan, i.e., data containing 5D information; Step 5) The data processing module performs filtered back-projection calculation processing on the 5D data to obtain the confidence value of each voxel as the surface information of the object outside the field of view corresponding to each voxel. This information can be visualized by the data processing module through a heat map and output as a reconstructed image.

2. The non-line-of-sight imaging simulation system and reconstruction method according to claim 1, characterized in that: The simulation system simulates a specific achievable non-line-of-sight imaging scenario; By emitting pulsed lasers to laser scanning points on the relay diffuse reflection wall, signals for reconstructing imaging of target objects hidden outside the field of view are obtained from the sensing points of the relay diffuse reflection wall; According to the scanning detection method, the information about the propagated distance light is accurately determined based on the use of high-speed captured photons, and this information is used to infer hidden geometric structures; the method is based on the ideas of energy front propagation and tomography, constructs forward and inversion models of cubic diffuse scattering, and restores the three-dimensional shape of non-field of view objects based on tomography methods and sparsity-based methods; the algorithm of the method is based on Fresnel approximation, simplifies the problem, and can achieve inversion through the back propagation process; the method can be used to discover and reconstruct hidden geometric structures.

3. The non-line-of-sight imaging simulation system and reconstruction method according to claim 1, characterized in that: The inversion and reconstruction imaging result of the non-viewing three-dimensional target object is an intuitive shape image of the non-viewing three-dimensional target reconstructed based on the data obtained by scanning, collecting and processing based on a filtered back-projection algorithm.

4. The non-line-of-sight imaging simulation system and reconstruction method according to claim 1, characterized in that: The laser pulse is generated by a narrow line width pulse laser, and the detection signal is obtained on the relay wall through regular dot scanning at certain time intervals, using a high time resolution device such as a streak camera or a single photon avalanche photodiode and its array as a sensor.

5. The method according to claim 1, characterized in that The reconstructed signal is obtained by confocal or non-confocal detection.

6. The non-line-of-sight imaging simulation system and reconstruction method according to claim 1, characterized in that: Both the galvanometer scanner and the camera are controlled by a computer.

7. The non-line-of-sight imaging simulation system and reconstruction method according to claim 1, characterized in that: The signal propagation process is simulated by a model of multiple reflection propagation of light pulses, and the model content includes: 1) A scene contains a hidden object, which is located in front of a first visible surface. The first surface may be a wall or a floor. The model assumes that the first visible surface (hereinafter referred to as the wall) is flat and diffusely reflective and follows the Lambertian reflector characteristics. The method aims to reconstruct the surface information of the object. 2) The laser is pointed at the wall. The pulsed laser emits very short light pulses. The light reflected by the wall reaches the hidden surface, is reflected and returns to the wall. The streak camera or single photon avalanche diode camera pointed at the wall records the time-varying image on the wall with very high time resolution. 3) For each position of the laser on the wall, the camera records a 3D image (2 spatial dimensions and 1 time dimension), and the laser point is moved to multiple positions on the wall (2D) to record 5D light transport data; 4) In the simulation case where the hidden scene is a single point, the reflected spherical wavefront propagating from the point arrives at different positions on the wall surface at different times, thereby generating a hyperbolic fringe image. There is a one-to-one reversible mapping between the parameters of the hyperbola and the three-dimensional position of the hidden scene point. When the hidden scene contains a surface instead of a single scene point, the spatiotemporal hyperbolas corresponding to different surface points are added together to produce the captured fringe image. The signal needs to be processed by a demultiplexing or deconvolution method; 5) There are four light path segments in the model: the light beam reaches the first surface, the light beam reaches the hidden surface point, the light beam reaches the first surface again, and finally reaches the camera; 6) The first and fourth optical paths are directional and focused, involving no scattering; the second optical path can be regarded as originating from an unfocused sensor, and the third optical path can be regarded as recorded by an unfocused sensor. By introducing the corresponding spatial and temporal distortions caused by the first and fourth optical paths, the model simplifies the three beam bounce problem into one beam bounce problem, similar to an unfocused emitter array and an unfocused receiver array. The goal of the method is to recover the position of each emitter from the multiplexed signal.

8. The non-line-of-sight imaging simulation system and reconstruction method according to claim 1, characterized in that: The imaging result of the non-viewing three-dimensional object is a main view of the non-viewing three-dimensional object obtained according to the filtered back-projection algorithm.

9. The non-line-of-sight imaging simulation system and reconstruction method according to claim 1, characterized in that: 1) When considering the influence of any scene point on the recorded 5D light transmission, the laser Illuminate a point on the visible surface of a wall , generating a virtual unfocused pulse light source from each scene point The reflected light is unfocused by the virtual sensor The model uses the Huygens-Fresnel principle to model the propagation of a pulse wave from each scene point, with the beam intensity being proportional to the total plane cross-section of the wavefront. changes, producing hyperbolic fringes, or simply "fringes", which are captured by a streak camera or a single-photon avalanche camera; 2) For the light scattering in the second and third paragraphs of the model, the hidden object is modeled as Time to send pulse The receiver is modeled as an array of unfocused receivers that capture photons with high temporal resolution. , the modeling can be achieved by using a high temporal resolution camera focused on the wall pixels. To simplify further, the model assumes that the speed of light is constant and expresses other quantities in distance units. The following mathematical equations represent the data recorded by the high temporal resolution camera after making the above assumptions: in , , , , represents the time shift between the transmitter and the receiver, and represents the intensity observed at a certain sensor at a certain moment.

10. The non-line-of-sight imaging simulation system and reconstruction method according to claim 1, characterized in that: When considering the positional relationship between the virtual unfocused transmitter array and the virtual unfocused receiver array separately, for a fixed transmitter position, the response function is a hyperboloid, and the spatial and temporal relationship of the hyperboloid is given by: . Where u, v are the position coordinates of the receiver on the receiver array plane.

11. The non-line-of-sight imaging simulation system and reconstruction method according to claim 1, characterized in that: If the position of the transmitter is fixed (or the position of the laser is fixed) and the position of the receiver, this equation is an ellipsoid at the transmitter's position, and an ellipse if the observation is restricted to a plane. Under the previous simplified conditions, the equation can be rewritten as: , this problem is similar to fault projection, and formula (3) can be rewritten as The world volume W(x) of the hidden object is a delta function with support on the reconstructed target surface, and the resulting single fringe image is the result of measuring the elliptical projection of the world volume W(x). This method is also called elliptical tomography by researchers.

12. The non-line-of-sight imaging simulation system and reconstruction method according to claim 1, characterized in that: For each voxel in the working volume and each fringe image, calculate the fringe image pixels that the voxel under consideration can contribute. Specifically, a voxel at position v can contribute to the pixel corresponding to point w on the wall at time t if Where C is the projection center of the camera and L is the laser position as described above. Stripe image pixels that meet this condition are called contributing pixels. The function and heat map on the voxel space are calculated to obtain the reconstructed image, as shown in Figures 2 and 3.

Images