Passenger head movement track reproduction method based on binocular camera and machine learning

Abstract

The invention provides a passenger head movement track reproduction method based on a binocular camera and machine learning. The method comprises the steps of importing a collision test video, performing head contour labeling based on the collision test video, performing model training and precision verification by using head contour labeling data through open source software, and completing a head contour capture model; acquiring a head contour and performing position calibration by using a binocular camera; when the head is not shielded by the airbag, acquiring pixel point coordinates of each frame of head contour, and calculating physical three-dimensional coordinates based on the pixel point coordinates; when the head is shielded by the airbag, acquiring pixel point coordinates of each frame of visual head contour, calculating the overall head contour form by using the head contour capture model, calculating pixel point coordinates of the overall head contour based on the pixel point coordinates of the visual head contour, and converting the pixel point coordinates into physical three-dimensional coordinates; when the head is rebounded, constructing a head contour movement track, and calculating the contact depth of the head and the air bag; and judging whether the head punctures the airbag.

Description

Technical Field

[0001] This application relates to the field of vehicle safety technology, specifically to a method for reproducing the occupant's head movement trajectory based on a binocular camera and machine learning. Background Technology

[0002] The method for assessing whether a dummy's head has been "penetrated" in sliding table or whole vehicle testing is to indirectly evaluate it using sensor data and animation. For example, if a "peak" suddenly appears in the head acceleration, and the overlap area between the head and the airbag in the video is used to "feel" that a head "penetration" has occurred, it is actually possible for a "small peak" to suddenly appear in the dummy's head during the test. In this case, it is difficult to determine whether a "penetration" has occurred, which requires engineers to add extra test resources to verify the test. The process is complex and costly. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the purpose of this application is to provide a method for reproducing the motion trajectory of an occupant's head based on a binocular camera and machine learning. This method can capture the head contour of a collision dummy based on collision video, establish head contour coordinates based on binocular ranging, and then depict the motion coordinates of the head in each frame to form a complete motion trajectory.

[0004] According to the method for reproducing occupant head motion trajectory based on binocular cameras and machine learning provided in this application, the method includes:

[0005] S1: Import the collision test video, perform head contour annotation based on the collision test video, use open source software to train the model with the head contour annotation data and verify the accuracy, and complete the head contour capture model.

[0006] S2: Obtain the head contour and use a binocular camera to perform position calibration;

[0007] S3: When the head is not obscured by the airbag, obtain the pixel coordinates of the head outline in each frame and calculate the physical 3D coordinates based on the pixel coordinates; when the head is obscured by the airbag, obtain the pixel coordinates of the visible head outline in each frame, use the head outline capture model to infer the overall head outline shape, infer the pixel coordinates of the overall head outline based on the pixel coordinates of the visible head outline, and convert them into physical 3D coordinates.

[0008] S4: After the head is bounced back, construct the head contour movement trajectory and calculate the contact depth between the head and the airbag;

[0009] S5: Determine if the head has been punctured by the airbag.

[0010] Preferably, the open-source software is YOLO X.

[0011] Preferably, the step of position calibration using a binocular camera in step S2 is as follows:

[0012] S21: Capture a picture of the checkerboard pattern and obtain corner images;

[0013] S22: Obtain the intrinsic and extrinsic parameters of the binocular camera by calculating the corner points of the checkerboard image;

[0014] S23: Perform stereoscopic correction and stereoscopic matching;

[0015] S24: Check the calculation results and verify them against the three-dimensional coordinate data.

[0016] Compared with existing technologies, this application provides a method for reproducing the occupant head motion trajectory based on binocular cameras and machine learning. It introduces binocular cameras, accurately calculates the object distance through the principle of triangle similarity, and uses machine learning to automatically capture the outline of the human head, jointly confirming the head trajectory in the collision scene and accurately judging the risk of injury. Attached Figure Description

[0017] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0018] Figure 1 This is a flowchart of the method for reproducing the occupant's head movement trajectory based on binocular cameras and machine learning in the embodiment. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0020] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] It should be noted that similar reference numerals and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, all directional indications in this application (such as up, down, left, right, front, back, bottom, etc.) are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indication will also change accordingly. Furthermore, descriptions involving "first," "second," etc., in this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0022] like Figure 1 As shown, this embodiment provides a method for reproducing occupant head motion trajectories based on binocular cameras and machine learning. The method includes:

[0023] S1: Import the collision test video, perform head contour annotation based on the collision test video, use YOLO X open source software to train the model with the head contour annotation data and verify the accuracy, and complete the head contour capture model.

[0024] S2: Obtain the head contour and perform position calibration using a binocular camera. Specifically: capture a checkerboard image and obtain corner point images; calculate the camera's internal and external parameters based on the corner points of the checkerboard image; perform stereo correction and stereo matching; and finally check the calculation results and verify them against the 3D coordinate data.

[0025] S3: When the head is not obscured by the airbag, obtain the pixel coordinates of the head outline in each frame and calculate the physical 3D coordinates based on the pixel coordinates; when the head is obscured by the airbag, obtain the pixel coordinates of the visible head outline in each frame, use the head outline capture model to infer the overall head outline shape, infer the pixel coordinates of the overall head outline based on the pixel coordinates of the visible head outline, and convert them into physical 3D coordinates.

[0026] S4: After the head is bounced back, construct the head contour movement trajectory and calculate the contact depth between the head and the airbag.

[0027] S5: Determine if the head has been punctured by the airbag.

[0028] The specific embodiments of this application have been described above. Based on the above description, those skilled in the art can make various changes and modifications without departing from the technical concept of this application.

Claims

1. A method for reproducing head movement trajectory of an occupant based on binocular camera and machine learning, characterized in that, The method comprises: S1: importing a collision test video, annotating a head contour based on the collision test video, training a model using open source software based on the head contour annotation data and verifying the accuracy, and completing a head contour capture model; S2: obtaining the head contour and calibrating the position using a binocular camera; S3: when the head is not blocked by the airbag, obtaining the pixel point coordinates of each frame of the head contour and calculating the physical three-dimensional coordinates based on the pixel point coordinates; when the head is blocked by the airbag, obtaining the pixel point coordinates of each frame of the visible head contour, calculating the overall head contour shape using the head contour capture model, calculating the pixel point coordinates of the overall head contour based on the pixel point coordinates of the visible head contour, and converting the pixel point coordinates into physical three-dimensional coordinates; S4: when the head is bounced back, constructing a head contour motion trajectory and calculating the contact depth of the head and the airbag; S5: judging whether the head has penetrated the airbag.

2. The method of claim 1, wherein the method comprises: The open source software is YOLO X.

3. The method of claim 1, wherein the method further comprises: The step of calibrating the position using the binocular camera in step S2 is: S21: taking a picture of a checkerboard and obtaining a corner image; S22: obtaining the internal and external parameters of the binocular camera based on the corner calculation of the checkerboard picture; S23: performing stereoscopic correction and stereoscopic matching; S24: checking the calculation result and checking it with three-dimensional coordinate data.

Images