Method for measuring distance between dummy head and vehicle-mounted screen based on multi-camera perspective

Abstract

The present application relates to the technical field of vehicle collision safety test, and particularly relates to a dummy head and vehicle screen distance measurement method based on multiple camera perspectives, comprising the following steps: S1: arranging multiple marker feature points on the dummy head, setting a reference point on the vehicle screen, and recording initial three-dimensional coordinates of each marker feature point and fixed three-dimensional coordinates of the vehicle screen; S2: deploying high-speed cameras and calibrating; S3: performing a simulated collision test, synchronously collecting image data and acceleration sensor data captured by multiple high-speed cameras; S4: extracting feature points of each frame of image and two-dimensional pixel coordinates of the marker feature points under each camera perspective; according to the number of available feature points of the current frame, calculating three-dimensional coordinates of each marker feature point of the current frame; S5: based on the three-dimensional coordinates of each marker feature point of the current frame and the fixed three-dimensional coordinates of the vehicle screen reference point, calculating the distance between the dummy head surface and the vehicle screen.

Description

Technical Field

[0001] This invention relates to the field of vehicle collision safety testing technology, and in particular to a method for measuring the distance between a dummy's head and an in-vehicle screen based on multiple camera perspectives. Background Technology

[0002] In vehicle crash safety testing, with the popularization of smart cockpit technology, large central control screens and passenger entertainment screens have gradually become mainstream configurations. These screens are usually installed in the center of the dashboard or on the passenger side, close to the area in front of the occupant's head. In frontal or offset collisions, the dummy's head may move forward due to inertia and collide with the in-vehicle screen. Unlike traditional steering wheels or dashboard structures, in-vehicle screens are mostly made of rigid glass, and their edge frames lack energy-absorbing designs. Therefore, accurately measuring the minimum distance between the dummy's head and the in-vehicle screen during a collision is of significant engineering importance for assessing the potential injury risk of the screen structure to occupants and for establishing standards for screen installation strength and crumple zone energy absorption design. The distance measurement results can be directly used to determine whether the occupant's head will come into contact with the screen under the action of existing restraint systems, as well as the relative velocity and energy magnitude at the time of contact, thus providing a quantitative basis for the placement, structural strength, and safety certification of in-vehicle screens.

[0003] In the existing technology, the methods for measuring the distance between the dummy's head and the vehicle screen are mainly divided into two categories, each with its own inherent defects.

[0004] The first type is based on accelerometer integration. This method uses accelerometers and angular velocity meters installed at the center of mass of the dummy's head to perform a second integration of the signals to calculate the head's trajectory. Its drawbacks include: accumulated error issues, and the inability to directly measure the actual distance between specific points on the head surface (such as the forehead or nose tip) and the screen surface, as the risk of contact between these points is often higher than at the center of mass. This results in a significant error in the calculated head-screen distance and fails to reflect the true risk of contact between the head surface and the screen, thus failing to provide accurate collision location information for screen collapse and energy absorption design.

[0005] The second category is based on high-speed vision methods. This method calculates displacement by attaching markers to the dummy's head and tracking their movement. However, this method faces unique challenges when measuring the distance between the head and the screen: First, the airbag deployment covers the area in front of the dummy's head, obscuring the head markers; second, the in-vehicle screen itself is made of high-brightness, highly reflective glass, and during a collision, the screen may experience changes in reflection due to vibration or breakage, leading to false feature points or mismatched markers in the camera image; third, the screen is usually mounted between the camera and the dummy's head, and the screen's frame and glass panel may partially obstruct the camera's view of the markers below the head, further reducing the number of available feature points. These factors cause the number of available feature points to fluctuate drastically over time during the measurement process using vision-based methods. When some markers are lost, methods relying solely on vision tracking cannot output reliable results. This results in gaps in the measurement data during the critical period after airbag deployment, making it impossible to obtain a complete head-screen distance change curve and making it difficult to assess the moment and speed of contact between the screen and the head. Summary of the Invention

[0006] This invention provides a method for measuring the distance between a dummy head and an in-vehicle screen based on multiple camera perspectives, which solves the problems of data gaps and insufficient accuracy in head-to-screen distance measurement caused by occlusion and the lack of adaptability of a single measurement method in the prior art.

[0007] The basic solution provided by this invention is a method for measuring the distance between a dummy's head and an in-vehicle screen based on multiple camera perspectives, comprising the following steps: S1: Arrange multiple marker feature points on the dummy's head, set reference points on the vehicle screen, and record the initial three-dimensional coordinates of each marker feature point in the vehicle coordinate system, as well as the fixed three-dimensional coordinates of the reference point on the vehicle screen in the vehicle coordinate system. S2: Deploy multiple high-speed cameras inside the test vehicle, calibrate them, and establish observation models for each camera; S3: Perform simulated collision tests, simultaneously acquiring image data from multiple high-speed cameras and acceleration sensor data; S4: Extract feature points from each frame of the image, obtain the two-dimensional pixel coordinates of the marked feature points from each camera viewpoint in the current frame, and count the number of available feature points in the current frame; based on the number of available feature points in the current frame, calculate the three-dimensional coordinates of each marked feature point in the current frame according to a preset motion trajectory estimation strategy; wherein, the preset motion trajectory estimation strategy is: When the number of available feature points under the view of a single camera meets the preset threshold, based on the observation model of that camera, the three-dimensional coordinates of each feature point in the current frame are calculated according to the initial three-dimensional coordinates and two-dimensional pixel coordinates of the available feature points in the current frame. When the number of available feature points from multiple camera perspectives meets the preset threshold, the three-dimensional coordinates of each feature point in the current frame are jointly calculated based on the observation model of each camera, the image data of multiple cameras are fused. When the number of available feature points under all camera views is lower than a preset threshold, the head displacement of the current frame is calculated based on the quadratic integral result of the accelerometer data, and the three-dimensional coordinates of each feature point in the current frame are calculated based on the head displacement. S5: Calculate the distance between the dummy's head surface and the vehicle screen based on the three-dimensional coordinates of each marked feature point in the current frame and the fixed three-dimensional coordinates of the reference point on the vehicle screen.

[0008] Preferably, in step S1, the strategy for arranging the feature points of the dummy head is as follows: Based on the geometry of the dummy's head and the coverage of multiple camera views, feature points are evenly distributed. The spacing between each feature point is determined based on the camera resolution and target tracking accuracy.

[0009] More preferably, in step S2, the deployment strategy for the high-speed camera is as follows: High-speed cameras were deployed inside the test vehicle, covering the area in front of, behind, and to the sides of the dummy. The field of view of the high-speed camera simultaneously covers the area of ​​motion of the dummy's head; There are at least 8 feature points within each initial camera viewpoint.

[0010] Preferably, in step S2, the strategy for constructing the camera's observation model is as follows: By capturing images of the calibration board, the intrinsic parameter matrix K of each camera is solved. The intrinsic parameter matrix includes focal length, principal point coordinates, and distortion coefficients. Taking one camera as the reference camera, the rotation matrix and translation vector of the other cameras relative to the reference camera are solved as the extrinsic parameters of each camera. The intrinsic parameter matrix and the extrinsic parameter matrix together constitute the projection model of each camera.

[0011] Preferably, in step S4, the preset threshold is determined based on the camera projection model and the motion trajectory estimation algorithm.

[0012] Preferably, in step S4, when the number of available feature points under only a single camera viewpoint meets a preset threshold, the three-dimensional coordinates of each feature point in the current frame are calculated in the following way: The initial 3D coordinates of the available feature points in the current frame are converted into theoretical pixel coordinates using the camera's projection model; Calculate the reprojection error between the theoretical pixel coordinates and the two-dimensional pixel coordinates extracted from the current frame; The first loss function is constructed using the sum of squares of the reprojection errors; The rotation matrix and translation vector of the current frame header are iteratively adjusted using a nonlinear optimization algorithm. Based on the optimized rotation matrix and translation vector, calculate the three-dimensional coordinates of each feature point in the current frame.

[0013] More preferably, the first loss function is:

[0014] in, For the camera The first frame of the image The pixel coordinates of each feature point Let P be the rotation matrix of the camera in the j-th frame. Let j be the translation vector of the camera in the j-th frame. Let the camera depth be the j-th frame. By using a sliding window to perform nonlinear optimization on the camera intrinsic parameter matrix based on the aforementioned least squares problem, the motion trajectory of the dummy head can be effectively estimated. P represents the coordinates of the marked feature point.

[0015] Preferably, in step S4, when the number of available feature points from multiple camera perspectives meets a preset threshold, the three-dimensional coordinates of each feature point in the current frame are calculated in the following way: The available feature points extracted from each camera viewpoint are correlated across viewpoints, and different camera observations belonging to the same spatial feature point are grouped into the same observation group. The initial 3D coordinates of the available feature points in the current frame are converted into theoretical pixel coordinates from the viewpoint of each camera using the projection model of each camera. Calculate the reprojection error between the theoretical pixel coordinates and the extracted two-dimensional pixel coordinates under each camera viewpoint; The joint loss function is constructed by summing the squares of the reprojection errors from the camera's viewpoint; The rotation matrix and translation vector of the current frame header are iteratively adjusted using a nonlinear optimization algorithm. Based on the optimized rotation matrix and translation vector, calculate the three-dimensional coordinates of each feature point in the current frame.

[0016] More preferably, the joint loss function is:

[0017] in, For the first The camera The first frame of the image The pixel coordinates of each feature point Where N is the number of cameras, M is the number of feature points optimized by a single camera, and M is the number of frames in a single camera's trial. and This is the rotation matrix and translation vector of the relative pose from the k-th camera to the first camera (used for relative coordinate transformation between the two cameras). Let be the rotation matrix of the i-th feature point in the j-th frame relative to the first frame within the first camera coordinate system. For the corresponding translation vector, for The camera depth of the point, where P is the coordinate of the marked feature point.

[0018] Preferably, when the number of available feature points from all camera views is lower than a preset threshold, the three-dimensional coordinates of each feature point in the current frame are calculated in the following way: The acceleration sensor data in the X, Y, and Z directions are integrated twice to obtain the displacement components of the current frame header in the three directions. The initial three-dimensional coordinates of the available feature points in the current frame are added to the displacement components in the corresponding directions to obtain the three-dimensional coordinates of each feature point in the current frame.

[0019] The principles and advantages of this invention are as follows: 1. Based on the actual number of usable feature points in each frame, three motion estimation strategies are dynamically switched. Compared with existing technologies, this approach covers complex scenarios such as airbag deployment, screen vibration and damage, and partial occlusion. It can employ high-precision visual estimation when feature points are sufficient, maintain estimation using multi-camera redundancy information when some feature points are lost, and switch to sensor short-time integral compensation when all feature points are lost, ensuring that usable distance data is always available throughout the entire collision process. This method is applicable not only to the entire collision scenario but also to real-world driving scenarios.

[0020] 2. This invention measures the real-time distance between a specific point on the dummy's head and the vehicle screen by placing marked feature points on the dummy's head surface, rather than relying on indirect calculations from a head center of mass sensor. This measurement method is closer to the real scenario of the head surface contacting the screen in an actual collision, and can provide more accurate collision location and contact velocity information for screen collapse energy absorption design and occupant injury assessment.

[0021] 3. This invention adopts a multi-camera perspective arrangement scheme, deploying high-speed cameras in front, on the side and behind the dummy to form a multi-view redundant observation system. When a camera perspective is blocked by an airbag or vehicle screen, other camera perspectives can still continue to capture feature points, effectively avoiding the measurement interruption problem caused by occlusion under a single camera perspective.

[0022] Preferably, when using image data for motion estimation, multi-camera and sensor information are jointly optimized. A least squares problem is established through reprojection error, and a nonlinear optimization iterative method is used in conjunction with sensor trajectory prediction results to improve prediction accuracy. Attached Figure Description

[0023] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0024] The following detailed description illustrates the specific implementation method: The specific implementation process is as follows: (See details) Figure 1 A method for measuring the distance between a dummy's head and an in-vehicle screen based on multiple camera perspectives includes the following steps: S1: Arrange multiple marker feature points on the dummy's head, set reference points on the vehicle screen, and record the initial three-dimensional coordinates of each marker feature point in the vehicle coordinate system, as well as the fixed three-dimensional coordinates of the reference point on the vehicle screen in the vehicle coordinate system. In step S1, the strategy for arranging the feature points of the dummy head is as follows: Based on the geometry of the dummy head and the coverage of the multi-camera viewpoints, feature points are evenly distributed. The distribution areas include the left and right head centroids, the center of the forehead, the area above the left and right brow ridges, the tip of the nose, the left and right cheekbones, the center of the chin, the left and right mandibular angles, the top of the head, and the back of the head, so that the feature points cover the entire dummy head. These areas can cover the main outer surface of the dummy head and have good visibility in different collision directions. The spacing between feature points is determined based on camera resolution and target tracking accuracy. The minimum spacing between adjacent feature points is set to a critical distance that can be distinguished by the camera's optical system and avoids feature point overlap in the image. In this embodiment, a gap of at least 30mm is guaranteed between feature points. Assuming the camera resolution is W×H (pixels), the field of view is θ, and the distance from the camera to the dummy's head is L, the spatial dimension corresponding to each pixel is approximately δ=(2Ltan(θ / 2)) / W. To ensure that two adjacent feature points can be clearly distinguished in the image (avoiding pixel overlap), the actual spatial spacing between feature points should be at least greater than 3δ. In this embodiment, based on camera parameters (W=1024 pixels, θ=30°, L≈1.5m), δ≈0.8mm is calculated. Therefore, the minimum spacing is set to 30mm, which is much greater than 3δ≈2.4mm, ensuring stable feature point tracking.

[0025] The spatial distribution density of each feature point is non-uniformly optimized based on the visibility constraints under each camera view. The feature point density is increased in areas that are easily occluded under the front camera view, and the feature point density is reduced in the inherently visible areas under the side camera view, so as to ensure that the number of available feature points is balanced under each camera view and is not lower than the preset threshold.

[0026] The reference points for the vehicle screen are set as the four corner points of the vehicle screen, denoted as follows: , , , The four corner points form a spatial rectangle, representing the effective area of ​​the vehicle screen.

[0027] S2: Deploy multiple high-speed cameras inside the test vehicle, calibrate them, and establish observation models for each camera; In step S2, the deployment strategy for the high-speed camera is as follows: Based on the spatial distribution characteristics of the dummy's head movement trajectory and the installation position of the vehicle-mounted screen, high-speed cameras were deployed inside the test vehicle, covering the area in front of, behind, and to the sides of the dummy. Specifically, the high-speed camera directly in front of the dummy was positioned so that the center line of the camera's frontal view coincided with the center line of the dummy's head. The high-speed camera on the left side of the dummy was positioned between the A and B pillars, at the same height as the dummy's chest, to ensure that the dummy's head and the vehicle-mounted screen were clearly visible during the test. The high-speed camera on the right side of the dummy was symmetrical to the one on the left. The high-speed camera at the rear of the dummy was located at the top rear to ensure that the dummy's head was clearly visible during the contact between the airbag and the head. The high-speed camera's field of view simultaneously covers the dummy's head movement area and the vehicle's screen area; There are at least 8 feature points within each initial camera viewpoint; Furthermore, the sampling frequency of each camera is determined based on the maximum angular velocity and linear velocity of the dummy head during the collision process, to ensure that the pixel displacement of feature points between two adjacent frames does not exceed the preset maximum tracking window size; the cameras achieve frame-level synchronization through hardware trigger signals, and the synchronization accuracy is not lower than the single frame exposure time. In this embodiment, a frequency of 2000 fps is used.

[0028] In step S2, the strategy for constructing the camera's observation model is as follows: 1) By capturing images of the calibration board, the intrinsic parameter matrix of each camera is solved. The intrinsic parameter matrix includes focal length, principal point coordinates, and distortion coefficients. 2) Using one of the cameras as the reference camera, solve for the rotation matrix and translation vector of the other cameras relative to the reference camera, and use them as the extrinsic parameters of each camera; The projection model of each camera is formed by the intrinsic parameter matrix and the extrinsic parameter matrix.

[0029] Specifically, the camera projection model during dummy movement: In the initial state (before the dummy head has moved), the coordinates of a certain feature point P on the dummy head are: The projection under a certain camera image frame should satisfy:

[0030] in, Let P be the depth of the camera model. Let P be the pixel coordinates of point P on the camera projection plane (camera image), and K be the camera intrinsic parameter; When the dummy's head moves, the projection of point P onto the corresponding image frame of the camera should satisfy:

[0031]

[0032] in, Let P be the depth of the camera model after the point P moves. Let P be the pixel coordinates of point P after its movement on the camera projection plane (camera image), K be the camera intrinsic parameter (the transformation matrix of point P's 3D coordinates after normalized projection, distortion correction, and conversion of camera intrinsic parameters into pixel coordinates), R be the rotation matrix of the head movement, and t be the translation vector of the head movement.

[0033] S3: Perform simulated collision tests, simultaneously acquiring image data from multiple high-speed cameras and acceleration sensor data; S4: Extract feature points from each frame of the image, obtain the two-dimensional pixel coordinates of the marked feature points from each camera viewpoint in the current frame, and count the number of available feature points in the current frame; based on the number of available feature points in the current frame, calculate the three-dimensional coordinates of each marked feature point in the current frame according to the preset motion trajectory estimation strategy. The preset trajectory estimation strategy is as follows: When the number of available feature points under the view of a single camera meets the preset threshold, based on the observation model of that camera, the three-dimensional coordinates of each feature point in the current frame are calculated according to the initial three-dimensional coordinates and two-dimensional pixel coordinates of the available feature points in the current frame. In step S4, when the number of available feature points from only a single camera viewpoint meets a preset threshold, the 3D coordinates of each feature point in the current frame are calculated in the following way: The initial 3D coordinates of the available feature points in the current frame are converted into theoretical pixel coordinates using the camera's projection model; Calculate the reprojection error between the theoretical pixel coordinates and the two-dimensional pixel coordinates extracted from the current frame; The first loss function is constructed using the sum of squares of the reprojection errors; The first loss function is:

[0034] in, For the camera The first frame of the image The pixel coordinates of each feature point Let P be the rotation matrix of the camera in the j-th frame. Let j be the translation vector of the camera in the j-th frame. Let the camera depth be the j-th frame. Let be the camera intrinsic parameter matrix. By using a sliding window to perform nonlinear optimization on the camera images that meet the requirements based on the above least squares problem, the motion trajectory of the dummy head can be effectively estimated. P represents the coordinates of the marked feature point; The rotation matrix and translation vector of the current frame header are iteratively adjusted using a nonlinear optimization algorithm. Based on the optimized rotation matrix and translation vector, calculate the three-dimensional coordinates of each feature point in the current frame.

[0035] When the number of available feature points from multiple camera perspectives meets the preset threshold, the three-dimensional coordinates of each feature point in the current frame are jointly calculated based on the observation model of each camera, the image data of multiple cameras are fused. In step S4, when the number of available feature points from multiple camera views meets a preset threshold, the 3D coordinates of each feature point in the current frame are calculated in the following way: The available feature points extracted from each camera viewpoint are correlated across viewpoints, and different camera observations belonging to the same spatial feature point are grouped into the same observation group. The initial 3D coordinates of the available feature points in the current frame are converted into theoretical pixel coordinates from the viewpoint of each camera using the projection model of each camera. Calculate the reprojection error between the theoretical pixel coordinates and the extracted two-dimensional pixel coordinates under each camera viewpoint; The joint loss function is constructed by summing the squares of the reprojection errors from the camera's viewpoint; The joint loss function is:

[0036] in, For the first The camera The first frame of the image The pixel coordinates of each feature point Where N is the number of cameras, M is the number of feature points optimized by a single camera, and M is the number of frames in a single camera's trial. and This is the rotation matrix and translation vector of the relative pose from the k-th camera to the first camera (used for relative coordinate transformation between the two cameras). Let be the rotation matrix of the i-th feature point in the j-th frame relative to the first frame within the first camera coordinate system. For the corresponding translation vector, for The camera depth of the point, where P is the coordinate of the marked feature point.

[0037] The rotation matrix and translation vector of the current frame header are iteratively adjusted using a nonlinear optimization algorithm. Based on the optimized rotation matrix and translation vector, calculate the three-dimensional coordinates of each feature point in the current frame.

[0038] When the number of available feature points under all camera views is lower than a preset threshold, the head displacement of the current frame is calculated based on the quadratic integral result of the accelerometer data, and the three-dimensional coordinates of each feature point in the current frame are calculated based on the head displacement. When the number of available feature points from all camera views is lower than a preset threshold, the 3D coordinates of each feature point in the current frame are calculated using the following method: The acceleration sensor data in the X, Y, and Z directions are integrated twice to obtain the displacement components of the current frame header in the three directions; the specific calculation formula is as follows:

[0039] In the formula, This represents the displacement component of the current frame header. The initial velocity of the head, For accelerometer data; The initial three-dimensional coordinates of the available feature points in the current frame are added to the displacement components in the corresponding directions to obtain the three-dimensional coordinates of each feature point in the current frame.

[0040] Preferably, in step S4, the preset threshold is determined based on the camera projection model and motion trajectory estimation algorithm. In this embodiment, the preset threshold is 10.

[0041] Specifically, based on the camera projection model established above:

[0042] Take the outer product of both sides of the equation with t:

[0043] Need to solve , is a 3x3 matrix, which requires at least 8 points to solve, so the preset threshold in this embodiment is 8.

[0044] S5: Calculate the distance between the dummy's head surface and the vehicle screen based on the three-dimensional coordinates of each marked feature point in the current frame and the fixed three-dimensional coordinates of the reference point on the vehicle screen.

[0045] The calculation strategy for the distance between the dummy's head surface and the vehicle screen is as follows: 5-1) Establish a spatial geometric model of the vehicle screen based on the fixed three-dimensional coordinates of the vehicle screen reference points; specifically, calculate the plane equation of the plane containing the vehicle screen based on the coordinates of any three vehicle screen reference points (corner points), denoted as:

[0046] In the formula, (A, B, C) are the plane normal vectors, and D is a constant term.

[0047] 5-2) Calculate the minimum distance from the marked feature point to the vehicle screen; a. Calculate the vertical distance from the marked feature points to the plane of the vehicle screen. Here, the three-dimensional coordinates of the i-th labeled feature point in the j-th frame image are represented as follows: ;

[0048] b. Calculate the projection points of the marked feature points onto the vehicle screen plane. The coordinates; c. Based on the projection point The coordinates are used to determine the distance between the dummy's head surface and the vehicle's screen; The judgment rule is: If the projection point Within the rectangle formed by the reference points on the vehicle screen, that point The minimum distance to the in-vehicle screen is this point. Distance to the in-vehicle screen plane ; If the projection point Outside the rectangle formed by the reference points of the vehicle screen, the projection point is calculated. The minimum value is taken from the distances to the four corner points (four reference points) of the vehicle screen and the distances to the four sides of the rectangle formed by the reference points. (As the shortest planar distance from the projection point to the rectangular region), this point Minimum distance to the in-vehicle screen According to the projection point Shortest planar distance to the rectangular area of ​​the vehicle screen and point To the projection point The distance is determined, i.e., the point Minimum distance to the in-vehicle screen For projection point Shortest planar distance to the rectangular area of ​​the vehicle screen With point To the projection point The distance to the hypotenuse; Specifically, point The minimum distance to the in-vehicle screen is:

[0049] 5-3) For each frame, calculate the minimum distance between all head-marked feature points in the current frame to obtain the minimum head-to-screen distance for that frame; finally, iterate through all frames and take the minimum value of the minimum distances between all marked feature points in all frames as the distance between the dummy's head and the vehicle screen throughout the entire experiment.

[0050] In the formula, For point Minimum distance to the in-vehicle screen.

[0051] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for measuring the distance between a dummy's head and a vehicle-mounted screen based on multiple camera perspectives, characterized in that, Includes the following steps: S1: Arrange multiple marker feature points on the dummy's head, set reference points on the vehicle screen, and record the initial three-dimensional coordinates of each marker feature point in the vehicle coordinate system, as well as the fixed three-dimensional coordinates of the reference point on the vehicle screen in the vehicle coordinate system. S2: Deploy multiple high-speed cameras inside the test vehicle, calibrate them, and establish observation models for each camera; S3: Perform simulated collision tests, simultaneously acquiring image data from multiple high-speed cameras and acceleration sensor data; S4: Extract feature points from each frame of the image, obtain the two-dimensional pixel coordinates of the marked feature points from each camera viewpoint in the current frame, and count the number of available feature points in the current frame. Based on the number of available feature points in the current frame, and according to a preset motion trajectory estimation strategy, the 3D coordinates of each marked feature point in the current frame are calculated; wherein, the preset motion trajectory estimation strategy is: When the number of available feature points under the view of a single camera meets the preset threshold, based on the observation model of that camera, the three-dimensional coordinates of each feature point in the current frame are calculated according to the initial three-dimensional coordinates and two-dimensional pixel coordinates of the available feature points in the current frame. When the number of available feature points from multiple camera perspectives meets the preset threshold, the three-dimensional coordinates of each feature point in the current frame are jointly calculated based on the observation model of each camera, the image data of multiple cameras are fused. When the number of available feature points under all camera views is lower than a preset threshold, the head displacement of the current frame is calculated based on the quadratic integral result of the accelerometer data, and the three-dimensional coordinates of each feature point in the current frame are calculated based on the head displacement. S5: Calculate the distance between the dummy's head surface and the vehicle screen based on the three-dimensional coordinates of each marked feature point in the current frame and the fixed three-dimensional coordinates of the reference point on the vehicle screen.

2. The method for measuring the distance between a dummy's head and an in-vehicle screen based on multiple camera perspectives as described in claim 1, characterized in that: In step S1, the strategy for arranging the feature points of the dummy head is as follows: Based on the geometry of the dummy's head and the coverage of multiple camera views, feature points are evenly distributed. The spacing between each feature point is determined based on the camera resolution and target tracking accuracy.

3. The method for measuring the distance between a dummy's head and an in-vehicle screen based on multiple camera perspectives as described in claim 2, characterized in that: In step S2, the deployment strategy for the high-speed camera is as follows: High-speed cameras were deployed inside the test vehicle, covering the area in front of, behind, and to the sides of the dummy. The field of view of the high-speed camera simultaneously covers the area of ​​motion of the dummy's head; There are at least 8 feature points within each initial camera viewpoint.

4. The method for measuring the distance between a dummy's head and an in-vehicle screen based on multiple camera perspectives according to claim 1, characterized in that: In step S2, the strategy for constructing the camera's observation model is as follows: By capturing images of the calibration board, the intrinsic parameter matrix K of each camera is solved. The intrinsic parameter matrix includes focal length, principal point coordinates, and distortion coefficients. Taking one camera as the reference camera, the rotation matrix and translation vector of the other cameras relative to the reference camera are solved as the extrinsic parameters of each camera. The intrinsic parameter matrix and the extrinsic parameter matrix together constitute the projection model of each camera.

5. The method for measuring the distance between a dummy's head and an in-vehicle screen based on multiple camera perspectives according to claim 4, characterized in that: In step S4, the preset threshold is determined based on the camera projection model and motion trajectory estimation algorithm.

6. The method for measuring the distance between a dummy's head and an in-vehicle screen based on multiple camera perspectives according to claim 1, characterized in that: In step S4, when the number of available feature points from only a single camera viewpoint meets a preset threshold, the 3D coordinates of each feature point in the current frame are calculated in the following way: The initial 3D coordinates of the available feature points in the current frame are converted into theoretical pixel coordinates using the camera's projection model; Calculate the reprojection error between the theoretical pixel coordinates and the two-dimensional pixel coordinates extracted from the current frame; The first loss function is constructed using the sum of squares of the reprojection errors; The rotation matrix and translation vector of the current frame header are iteratively adjusted using a nonlinear optimization algorithm. Based on the optimized rotation matrix and translation vector, calculate the three-dimensional coordinates of each feature point in the current frame.

7. The method for measuring the distance between a dummy's head and an in-vehicle screen based on multiple camera perspectives as described in claim 6, characterized in that: The first loss function is: in, For the camera The first frame of the image The pixel coordinates of each feature point Let P be the rotation matrix of the camera in the j-th frame. Let j be the translation vector of the camera in the j-th frame. Let the camera depth be in frame j. By using a sliding window to perform nonlinear optimization on the camera intrinsic parameter matrix based on the aforementioned least squares problem, the motion trajectory of the dummy head can be effectively estimated. P represents the coordinates of the marked feature point.

8. The method for measuring the distance between a dummy's head and an in-vehicle screen based on multiple camera perspectives according to claim 1, characterized in that: In step S4, when the number of available feature points from multiple camera views meets a preset threshold, the 3D coordinates of each feature point in the current frame are calculated in the following way: The available feature points extracted from each camera viewpoint are correlated across viewpoints, and different camera observations belonging to the same spatial feature point are grouped into the same observation group. The initial 3D coordinates of the available feature points in the current frame are converted into theoretical pixel coordinates from the viewpoint of each camera using the projection model of each camera. Calculate the reprojection error between the theoretical pixel coordinates and the extracted two-dimensional pixel coordinates under each camera viewpoint; The joint loss function is constructed by summing the squares of the reprojection errors from the camera's viewpoint; The rotation matrix and translation vector of the current frame header are iteratively adjusted using a nonlinear optimization algorithm. Based on the optimized rotation matrix and translation vector, calculate the three-dimensional coordinates of each feature point in the current frame.

9. The method for measuring the distance between a dummy's head and an in-vehicle screen based on multiple camera perspectives as described in claim 8, characterized in that: The joint loss function is: in, For the first The camera The first frame of the image The pixel coordinates of each feature point Where N is the number of cameras, M is the number of feature points optimized by a single camera, and M is the number of frames in a single camera's trial. and This is the rotation matrix and translation vector of the relative pose from the k-th camera to the first camera (used for relative coordinate transformation between the two cameras). Let be the rotation matrix of the i-th feature point in the j-th frame relative to the first frame within the first camera coordinate system. For the corresponding translation vector, for The camera depth of the point, where P is the coordinate of the marked feature point.

10. The method for measuring the distance between a dummy's head and an in-vehicle screen based on multiple camera perspectives according to claim 1, characterized in that: When the number of available feature points from all camera views is lower than a preset threshold, the 3D coordinates of each feature point in the current frame are calculated using the following method: The acceleration sensor data in the X, Y, and Z directions are integrated twice to obtain the displacement components of the current frame header in the three directions. The initial three-dimensional coordinates of the available feature points in the current frame are added to the displacement components in the corresponding directions to obtain the three-dimensional coordinates of each feature point in the current frame.

Images