A vision-based method for reconstructing the pose of a multi-rigid pedestrian model

Abstract

The present invention discloses a method for reconstructing the posture of a multi-rigid-body pedestrian model based on vision, comprising: extracting an image frame containing a pedestrian from an accident video; inputting the image frame into a 3D human morphology recognition model, extracting the posture parameters, body shape parameters and camera parameters of the pedestrian in the image frame; importing the extracted posture parameters, body shape parameters and camera parameters into an SMPL parameterized human body model, generating the human posture and joint points of the SMPL model; initializing the original posture and joint position of the Madymo multi-rigid-body pedestrian model, and matching the joint points at the same position of the Madymo multi-rigid-body pedestrian model and the SMPL model; importing the posture parameters of the SMPL model into the Madymo multi-rigid-body pedestrian model, generating a dummy model with a real pedestrian posture. The method of the present invention can directly generate the posture of the dummy model in real time from the video frame, which is helpful for the rapid reconstruction of a human-vehicle collision accident.

Description

Technical Field

[0001] The invention relates to the technical field of traffic accident safety, and in particular to a vision-based multi-rigid-body pedestrian model posture reconstruction method. Background Art

[0002] Pedestrians are a vulnerable group in traffic accidents. In order to reduce the damage caused by collisions to pedestrians, the study of human-vehicle collision accidents has become a key issue of concern. By collecting information from the accident scene and before and after the collision, and conducting statistical analysis and simulation reconstruction after data processing, a large amount of valuable human-vehicle collision data can be obtained. This can provide key reference data for government agencies to revise road traffic safety regulations and standards, provide technical support for automobile manufacturers to develop active and passive safety protection systems for vehicles, and provide in-depth data support for scientific research institutions to conduct research on accident mechanisms and accident prevention research; thereby promoting the development of new automotive safety technologies and improving the level of automotive safety performance, and ultimately achieving the goal of reducing accident injuries and avoiding casualties.

[0003] However, traffic accidents are rare data, difficult to collect and with inconsistent standards. In research, real human collision tests are not suitable, and the commonly used cadaver experiments, volunteer experiments, and physical dummy models in collision tests are too expensive. With the development of computer simulation technology, the use of digital dummy models to reproduce human-vehicle accidents is currently one of the most widely used, fast and effective methods.

[0004] When using simulation software to study human-vehicle collision accidents, if a data-rich and accurate pedestrian model is added, it can not only scientifically restore the real accident, but also improve the accuracy of the simulation results. At present, dummy model scaling is the fastest way to establish a dummy model in human-vehicle accident reconstruction. By scaling the human body parameters of the key parts of the dummy, dummy models of different heights and shapes can be obtained to solve the differences in accidents caused by different pedestrian characteristics in the human body database, but it cannot reconstruct the real pedestrian posture in the accident. Pedestrian posture is an important factor affecting collision injuries. Most studies are still based on qualitative analysis. Some establish common emergency postures or gaits of pedestrians through statistical data in the database, and analyze the influencing factors in the collision of different dummy models. This method is mostly used to study the impact of collision of human body parts and cannot be applied to whole body posture. Some methods manually adjust the dummy posture from the accident video to obtain a posture similar to that in the video, but this will bring a lot of manpower costs and cannot be developed and applied on a large scale. Summary of the invention

[0005] The present invention provides a vision-based multi-rigid-body pedestrian model posture reconstruction method to solve the technical problem of excessively high labor cost in the existing method of constructing a dummy model with pedestrian posture.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] A vision-based multi-rigid-body pedestrian model posture reconstruction method, comprising:

[0008] Select a video clip to be reconstructed from the accident video, input the video clip into the pedestrian detection network, obtain a video containing a pedestrian detection frame, extract frames from the video containing the pedestrian detection frame, and crop the image frame obtained after the extraction according to the pedestrian detection frame to obtain a new image frame containing the pedestrian;

[0009] The cropped image frames are input into a pre-trained 3D human morphology recognition model based on deep learning to extract the posture parameters, body shape parameters and camera parameters of the pedestrians in the image frames;

[0010] Import the extracted posture parameters, body shape parameters and camera parameters into the SMPL (Skinned Multi-PersonLinear) model to generate the human posture and joint points of the SMPL model;

[0011] Initialize the original posture and joint position of the Madymo multi-rigid-body pedestrian model, and match the same position joint points of the Madymo multi-rigid-body pedestrian model and the SMPL model;

[0012] The human body posture and joint points of the SMPL model are imported into the Madymo multi-rigid body pedestrian model with the original posture and joint position initialization and joint point matching, generating a dummy model with real pedestrian posture.

[0013] Optionally, the pedestrian detection network is a YOLO network or a Faster-RCNN network.

[0014] Optionally, the deep learning-based 3D human body shape recognition model is an HMR or SPIN model.

[0015] Optionally, the posture parameter range is -π to π; the body shape parameter is used to control the height, fatness and body proportions of the dummy model; and the camera parameters include rotation, translation and scaling.

[0016] Optionally, the SMPL model includes a male SMPL model and a female SMPL model;

[0017] The step of importing the extracted posture parameters, body shape parameters and camera parameters into the SMPL model includes:

[0018] The SMPL model is selected according to the gender of the pedestrian in the image frame. When the pedestrian is male, the extracted posture parameters, body shape parameters and camera parameters are imported into the male SMPL model; when the pedestrian is female, the extracted posture parameters, body shape parameters and camera parameters are imported into the female SMPL model.

[0019] Optionally, the SMPL model joints include head, neck, left clavicle, right clavicle, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left fingers, right fingers, upper spine, middle spine, lower spine, pelvis, left hip, right hip, left knee, right knee, left ankle, right ankle, left toes and right toes.

[0020] Optionally, the Madymo multi-rigid-body pedestrian model includes: a female Madymo multi-rigid-body pedestrian model, a male Madymo multi-rigid-body pedestrian model and a child Madymo multi-rigid-body pedestrian model.

[0021] Optionally, the joints of the Madymo multi-rigid-body pedestrian model include: head-neck, upper neck-lower neck, upper torso-lower neck, left shoulder, right shoulder, upper waist-upper torso, left elbow, right elbow, lower waist-upper waist, pelvis, left hip, right hip, left wrist, right wrist, left knee, right knee, left ankle and right ankle.

[0022] Optionally, the initializing the original posture and joint position of the Madymo multi-rigid-body pedestrian model, and matching the same position joint points of the Madymo multi-rigid-body pedestrian model and the SMPL model, includes:

[0023] In the Madymo software, by adjusting the angles of the hinge points of the head-neck, left shoulder, right shoulder, left elbow, right elbow, left hip, and right hip of the Madymo multi-rigid-body pedestrian model, the posture of the Madymo multi-rigid-body pedestrian model is made into a "T" shape, which is consistent with the initial posture of the SMPL model;

[0024] Match the joints of the Madymo multi-rigid pedestrian model to the joints of the same positions in the SMPL model; when matching joints, the head-neck, upper neck-lower neck, left shoulder, right shoulder, upper waist-upper trunk, left elbow, right elbow, lower waist-upper waist, pelvis, left hip, right hip, left wrist, right wrist, left knee, right knee, left ankle and right ankle in the Madymo multi-rigid pedestrian model correspond to the head, neck, left shoulder, right shoulder, lower spine, left elbow, right elbow, middle spine, pelvis, left hip, right hip, left finger, right wrist, left knee, right knee, left ankle and right ankle in the SMPL model respectively;

[0025] Based on the rotation direction of the SMPL model joint degrees of freedom, the rotation direction of the degrees of freedom of the joint points of the Madymo multi-rigid-body pedestrian model is redefined to make it consistent with the rotation direction of the degrees of freedom of the joint points of the SMPL model.

[0026] Optionally, the step of importing the human body posture and joint points of the SMPL model into the Madymo multi-rigid body pedestrian model that has completed the original posture and joint position initialization and joint point matching to generate a dummy model with a real pedestrian posture includes:

[0027] According to the joint point matching relationship between the Madymo multi-rigid-body pedestrian model and the SMPL model and the defined degree of freedom rotation of the joint points of the Madymo multi-rigid-body pedestrian model, the human body posture and joint point data of the SMPL model are imported into the Madymo multi-rigid-body pedestrian model to generate a dummy model with a real pedestrian posture.

[0028] On the other hand, the present invention further provides an electronic device, comprising a processor and a memory; wherein the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the above method.

[0029] In yet another aspect, the present invention further provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, and the instruction is loaded and executed by a processor to implement the above method.

[0030] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0031] The vision-based multi-rigid-body pedestrian model posture reconstruction method provided by the present invention processes pedestrians in real accident videos based on deep learning technology, extracts pedestrian posture and body shape feature parameters, and then combines the Madymo multi-rigid-body pedestrian model to reconstruct the pedestrian posture in the accident. This achieves the effect of directly generating the dummy model posture in real time from the video frame, which in turn helps to quickly reconstruct pedestrian-vehicle collision accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 1 is a schematic diagram of the execution flow of the method for reconstructing the posture of a multi-rigid-body pedestrian model based on vision provided by the first embodiment of the present invention;

[0034] Figure 2is a schematic diagram of the implementation process of the vision-based multi-rigid-body pedestrian model posture reconstruction method provided in the second embodiment of the present invention;

[0035] Figure 3 It is the joint point matching diagram of SMPL model and Madymo model; (a) is the SMPL initial model, (b) is the Madymo original model, and (c) is the matched Madymo model;

[0036] Figure 4 Schematic diagram of pedestrians in video images and dummy models at different stages; (a) is the pedestrian image frame, (b) is the generated SMPL model, and (c) is the final Madymo model posture. DETAILED DESCRIPTION

[0037] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0038] First embodiment

[0039] This embodiment provides a method for reconstructing the posture of a multi-rigid-body pedestrian model based on vision. The method can be implemented by an electronic device. The execution process of the method is as follows: Figure 1 As shown, the following steps are included:

[0040] S1, select a video clip to be reconstructed from the accident video, input the video clip into the pedestrian detection network, obtain a video containing a pedestrian detection frame, extract frames from the video containing the pedestrian detection frame, and crop the image frames obtained after the extraction according to the pedestrian detection frame to obtain a new image frame containing the pedestrian;

[0041] Among them, the pedestrian detection network is a YOLO network or a Faster-RCNN network.

[0042] S2, inputting the cropped image frame into a pre-trained 3D human morphology recognition model based on deep learning to extract the posture parameters, body shape parameters and camera parameters of the pedestrian in the image frame;

[0043] The 3D human body shape recognition model based on deep learning is HMR or SPIN model. HMR is an existing human body parameterized model and a corresponding parameter detection method.

[0044] The posture parameter range is -π to π radians; the body shape parameters are used to control the height, fatness and body proportions of the dummy model; the camera parameters include rotation, translation and scaling.

[0045] S3, importing the extracted posture parameters, body shape parameters and camera parameters into the SMPL parameterized human body model to generate the human body posture and joint points of the SMPL model;

[0046] Among them, the SMPL model includes the male SMPL model and the female SMPL model;

[0047] The step of importing the extracted posture parameters, body shape parameters and camera parameters into the SMPL model includes:

[0048] The SMPL model is selected according to the gender of the pedestrian in the image frame. When the pedestrian is male, the extracted posture parameters, body shape parameters and camera parameters are imported into the male SMPL model; when the pedestrian is female, the extracted posture parameters, body shape parameters and camera parameters are imported into the female SMPL model.

[0049] The joints of the SMPL model include head, neck, left clavicle, right clavicle, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left finger, right finger, upper spine, middle spine, lower spine, pelvis, left hip, right hip, left knee, right knee, left ankle, right ankle, left toe and right toe.

[0050] S4, initialize the original posture and joint position of the Madymo multi-rigid pedestrian model, and match the same position joint points of the Madymo multi-rigid pedestrian model and the SMPL model;

[0051] Among them, Madymo multi-rigid pedestrian models include: standard female multi-rigid digital dummy, medium-sized standard male multi-rigid digital dummy, large-size standard male multi-rigid digital dummy, 3-year-old standard child multi-rigid digital dummy, and 6-year-old standard child multi-rigid digital dummy.

[0052] The joints of the Madymo multi-rigid-body pedestrian model include: head-neck, upper neck-lower neck, upper torso-lower neck, left shoulder, right shoulder, upper waist-upper torso, left elbow, right elbow, lower waist-upper waist, pelvis, left hip, right hip, left wrist, right wrist, left knee, right knee, left ankle and right ankle.

[0053] Initializing the original posture and joint position of the Madymo multi-rigid-body pedestrian model, and matching the same position joint points of the Madymo multi-rigid-body pedestrian model and the SMPL model, includes:

[0054] In the Madymo software, by adjusting the angles of the hinge points of the head-neck, left shoulder, right shoulder, left elbow, right elbow, left hip, and right hip of the Madymo multi-rigid-body pedestrian model, the posture of the Madymo multi-rigid-body pedestrian model is made into a "T" shape, which is consistent with the initial posture of the SMPL model;

[0055] Match the joints of the Madymo multi-rigid pedestrian model to the joints of the same positions in the SMPL model; when matching joints, the head-neck, upper neck-lower neck, left shoulder, right shoulder, upper waist-upper trunk, left elbow, right elbow, lower waist-upper waist, pelvis, left hip, right hip, left wrist, right wrist, left knee, right knee, left ankle and right ankle in the Madymo multi-rigid pedestrian model correspond to the head, neck, left shoulder, right shoulder, lower spine, left elbow, right elbow, middle spine, pelvis, left hip, right hip, left finger, right wrist, left knee, right knee, left ankle and right ankle in the SMPL model respectively;

[0056] Based on the rotation direction of the SMPL model joint degrees of freedom, the rotation direction of the degrees of freedom of the joint points of the Madymo multi-rigid-body pedestrian model is redefined to make it consistent with the rotation direction of the degrees of freedom of the joint points of the SMPL model.

[0057] S5, import the posture parameters of the SMPL model into the Madymo multi-rigid-body pedestrian model that completes the original posture and joint position initialization and joint point matching, and generates a dummy model with a real pedestrian posture.

[0058] Among them, the human body posture and joint points of the SMPL model are imported into the Madymo multi-rigid-body pedestrian model to complete the original posture and joint position initialization and joint point matching. Specifically, according to the joint point matching relationship between the Madymo multi-rigid-body pedestrian model and the SMPL model determined in S4 and the degree of freedom rotation direction of the joint points of the Madymo multi-rigid-body pedestrian model, the human body posture and joint point data of the SMPL model are imported into the Madymo multi-rigid-body pedestrian model, thereby generating a dummy model with a real pedestrian posture.

[0059] The vision-based multi-rigid-body pedestrian model posture reconstruction method provided in this embodiment processes pedestrians in real accident videos based on deep learning technology, extracts pedestrian posture and body shape feature parameters, and then reconstructs the pedestrian posture in the accident in combination with the Madymo multi-rigid-body pedestrian model. This achieves the effect of directly generating the dummy model posture in real time from the video frame, which in turn helps to quickly reconstruct pedestrian-vehicle collision accidents.

[0060] Second embodiment

[0061] This embodiment provides a method for reconstructing the posture of a multi-rigid-body pedestrian model based on vision. The method can be implemented by an electronic device. The implementation process of the method is as follows: Figure 2 shown.

[0062] Specifically, the vision-based multi-rigid-body pedestrian model posture reconstruction method includes the following steps:

[0063] S1, obtain the collision accident video clip from the vehicle recorder or road monitoring equipment, input it into the YOLOv3 pedestrian detection network, and obtain the video containing the pedestrian detection frame; extract the frame of the output video by the algorithm, and crop the picture according to the pedestrian detection frame to obtain a new pedestrian image frame, such as Figure 4 As shown in (a) in .

[0064] S2, input the obtained pedestrian image frame into the pre-trained HMR 3D human body reconstruction network model, and output 72 posture parameters, 10 body shape parameters and 6 camera parameters of the pedestrian.

[0065] S3, import the obtained posture parameters, body shape parameters and camera parameters into the male or female SMPL model according to the pedestrian's gender, and generate the human posture and 24 joint points of the SMPL model, such as Figure 3 As shown in (b).

[0066] Among them, Figure 3 As shown in (a) in the figure, the SMPL parameterized human body model includes 24 key point positions, including the following joints: head 15, neck 12, left clavicle 13, right clavicle 14, left shoulder 16, right shoulder 17, left elbow 18, right elbow 19, left wrist 20, right wrist 21, left finger 22, right finger 23, upper spine 9, middle spine 6, lower spine 3, pelvis 0, left hip 1, right hip 2, left knee 4, right knee 5, left ankle 7, right ankle 8, left toe 10, right toe 11.

[0067] S4, there are differences in the number of joints and initial posture parameters between the SMPL model parameters and the Madymo multi-rigid-body pedestrian model (hereinafter referred to as the Madymo model). Therefore, it is necessary to match the postures and joints of the SMPL model and the Madymo model to make them consistent with the parameters; the details are as follows:

[0068] like Figure 3 As shown in (b), the Madymo model has 18 joints, including: head-neck a, upper neck-lower neck b, upper torso-lower neck c, left shoulder d, right shoulder e, upper waist-upper torso f, left elbow g, right elbow h, lower waist-upper waist i, pelvis j, left hip k, right hip l, left wrist m, right wrist n, left knee o, right knee p, left ankle q, right ankle r;

[0069] Dummy model initialization: Figure 3 As shown in (c), in the Madymo software, by adjusting the angles of the hinge points of the head-neck a, left shoulder d, right shoulder e, left elbow g, right elbow h, left hip k, and right hip l of the Madymo model, its posture is in a "T" shape, which is consistent with the initial posture of the SMPL model;

[0070] Joint point matching: Match the joint points of the Madymo pedestrian model with the joint points at the same position in the SMPL model, and establish the corresponding relationship between the joint points of the two: select 17 pairs of joint points: (a,15), (b,12), (d,16), (e,17), (f,3), (g,18), (h,19), (i,6), (j,0), (k,1), (l,2), (m,22), (n,21), (o,4), (p,5), (q,7), (r,8), and the rest of the unpaired joint points are not considered;

[0071] Initialization of posture parameters: Based on the rotation direction of the rotational freedom of the SMPL model joints, redefine the rotational freedom direction of the Madymo model joints to make it consistent with the rotational freedom direction of the SMPL model joints.

[0072] S5, import the posture parameters of the SMPL model into the adjusted Madymo model according to the above joint point matching method and the defined rotation direction, and generate a file to store the dummy information, as shown in the visualization. Figure 4 As shown in (c).

[0073] The vision-based multi-rigid-body pedestrian model posture reconstruction method provided in this embodiment processes pedestrians in real accident videos based on deep learning technology, extracts pedestrian posture and body shape feature parameters, and then reconstructs the pedestrian posture in the accident in combination with the Madymo multi-rigid-body pedestrian model. This achieves the effect of directly generating the dummy model posture in real time from the video frame, which in turn helps to quickly reconstruct pedestrian-vehicle collision accidents.

[0074] Third embodiment

[0075] This embodiment provides an electronic device, which includes a processor and a memory; wherein the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the method of the above embodiment.

[0076] The electronic device may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) and one or more memories, wherein the memory stores at least one instruction, and the instruction is loaded by the processor to execute the above method.

[0077] Fourth embodiment

[0078] This embodiment provides a computer-readable storage medium, which stores at least one instruction, and the instruction is loaded and executed by a processor to implement the method of the above embodiment. The computer-readable storage medium can be a ROM, a random access memory, a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device. The instructions stored therein can be loaded by a processor in a terminal to execute the above method.

[0079] In addition, it should be noted that the present invention can be provided as a method, an apparatus or a computer program product. Therefore, the embodiments of the present invention can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0080] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0081] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable terminal device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0082] It should also be noted that, in this article, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0083] Finally, it should be noted that the above is a preferred embodiment of the present invention. It should be pointed out that although the preferred embodiment of the present invention has been described, for those skilled in the art, once the basic creative concept of the present invention is known, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention. Therefore, the attached claims are intended to be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the embodiments of the present invention.

Claims

1. A vision-based multi-rigid-body pedestrian model posture reconstruction method, characterized in that: include: Select a video clip to be reconstructed from the accident video, input the video clip into the pedestrian detection network, obtain a video containing a pedestrian detection frame, extract frames from the video containing the pedestrian detection frame, and crop the image frame obtained after the extraction according to the pedestrian detection frame to obtain a new image frame containing the pedestrian; The cropped image frames are input into a pre-trained 3D human morphology recognition model based on deep learning to extract the posture parameters, body shape parameters and camera parameters of the pedestrians in the image frames; Import the extracted posture parameters, body shape parameters and camera parameters into the SMPL (Skinned Multi-Person Linear) model to generate the human posture and joint points of the SMPL model; Initialize the original posture and joint position of the Madymo multi-rigid-body pedestrian model, and match the same position joint points of the Madymo multi-rigid-body pedestrian model and the SMPL model; Import the human body posture and joint points of the SMPL model into the Madymo multi-rigid-body pedestrian model with the original posture and joint position initialization and joint point matching, and generate a dummy model with the real pedestrian posture; The joints of the SMPL model include head, neck, left clavicle, right clavicle, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left finger, right finger, upper spine, middle spine, lower spine, pelvis, left hip, right hip, left knee, right knee, left ankle, right ankle, left toe and right toe; The joints of the Madymo multi-rigid-body pedestrian model include: head-neck, neck-neck, torso-neck, left shoulder, right shoulder, waist-torso, left elbow, right elbow, waist-waist, pelvis, left hip, right hip, left wrist, right wrist, left knee, right knee, left ankle and right ankle. Initializing the original posture and joint position of the Madymo multi-rigid-body pedestrian model, and matching the same position joint points of the Madymo multi-rigid-body pedestrian model and the SMPL model, includes: In the Madymo software, by adjusting the angles of the hinge points of the head-neck, left shoulder, right shoulder, left elbow, right elbow, left hip and right hip of the Madymo multi-rigid-body pedestrian model, the posture of the Madymo multi-rigid-body pedestrian model is made into a "T" shape, which is consistent with the initial posture of the SMPL model; Match the joints of the Madymo multi-rigid pedestrian model to the joints of the same positions in the SMPL model; when matching joints, the head-neck, upper neck-lower neck, left shoulder, right shoulder, upper waist-upper trunk, left elbow, right elbow, lower waist-upper waist, pelvis, left hip, right hip, left wrist, right wrist, left knee, right knee, left ankle and right ankle in the Madymo multi-rigid pedestrian model correspond to the head, neck, left shoulder, right shoulder, lower spine, left elbow, right elbow, middle spine, pelvis, left hip, right hip, left finger, right wrist, left knee, right knee, left ankle and right ankle in the SMPL model respectively; Based on the rotation direction of the SMPL model joint degrees of freedom, the rotation direction of the degrees of freedom of the joint points of the Madymo multi-rigid body pedestrian model is redefined to make it consistent with the rotation direction of the degrees of freedom of the joint points of the SMPL model; The method of importing the human body posture and joint points of the SMPL model into the Madymo multi-rigid body pedestrian model that completes the original posture and joint position initialization and joint point matching, and generating a dummy model with a real pedestrian posture, includes: According to the joint point matching relationship between the Madymo multi-rigid-body pedestrian model and the SMPL model and the defined degree of freedom rotation of the joint points of the Madymo multi-rigid-body pedestrian model, the human body posture and joint point data of the SMPL model are imported into the Madymo multi-rigid-body pedestrian model to generate a dummy model with a real pedestrian posture.

2. The method for reconstructing the posture of a multi-rigid-body pedestrian model based on vision as claimed in claim 1, characterized in that: The pedestrian detection network is a YOLO network or a Faster-RCNN network.

3. The method for reconstructing the posture of a multi-rigid-body pedestrian model based on vision as claimed in claim 1, characterized in that: The 3D human body shape recognition model based on deep learning is an HMR or SPIN model.

4. The method for reconstructing the posture of a multi-rigid-body pedestrian model based on vision as claimed in claim 1, characterized in that: The range of the posture parameters is -π to π; the body shape parameters are used to control the height, fatness and body proportions of the dummy model; and the camera parameters include rotation, translation and scaling.

5. The method for reconstructing the posture of a multi-rigid-body pedestrian model based on vision as claimed in claim 1, characterized in that: The SMPL model includes a male SMPL model and a female SMPL model; The step of importing the extracted posture parameters, body shape parameters and camera parameters into the SMPL model includes: The SMPL model is selected according to the gender of the pedestrian in the image frame. When the pedestrian is male, the extracted posture parameters, body shape parameters and camera parameters are imported into the male SMPL model; when the pedestrian is female, the extracted posture parameters, body shape parameters and camera parameters are imported into the female SMPL model.

6. The method for reconstructing the posture of a multi-rigid-body pedestrian model based on vision as claimed in claim 1, characterized in that: The Madymo multi-rigid-body pedestrian model includes: a female Madymo multi-rigid-body pedestrian model, a male Madymo multi-rigid-body pedestrian model and a child Madymo multi-rigid-body pedestrian model.

Images