Search

CN-121994217-A - System and method for robust acquisition of 6D pose data of head of dummy in collision test

CN121994217ACN 121994217 ACN121994217 ACN 121994217ACN-121994217-A

Abstract

The invention relates to the technical field of data acquisition and analysis of automobile safety collision tests, in particular to a system and a method for robust acquisition of 6D gesture data of a head of a collision test dummy. The invention comprises an edge AI analysis module, a visual depth perception unit, a dummy head inertia measurement unit, a redundant power supply and data security module and an inertia wake-up and gesture reference unit. The invention realizes accurate 6D pose measurement, effectively solves the problem of interruption or inaccuracy of visual/inertial data caused by airbag shielding, environmental change and inertial drift through visual depth fusion and unscented Kalman filtering on the dummy head inertial measurement data, ensures the robustness and continuity of the 6D pose data and the head high-precision dynamic information in the whole collision process, and still has a second scheme as a bottom protection mechanism under extreme conditions to acquire the dummy head related data, thereby being capable of providing high-precision high-dimensional analysis data for automobile manufacturers.

Inventors

  • TUO JIYING
  • CHEN ZHENFENG
  • HUANG AN
  • LIU ZILIN
  • HU DU
  • WEI XIAOYU
  • QI WENJIE

Assignees

  • 重庆理工大学

Dates

Publication Date
20260508
Application Date
20251218

Claims (9)

  1. 1. The utility model provides a collision test dummy head 6D gesture data robust acquisition system which is characterized in that, includes edge AI analysis module, vision depth perception unit, dummy head inertial measurement unit and redundant power and data security module, the deployment has the unscented Kalman filtering fusion algorithm in the edge AI analysis module for fuse 6D gesture data that vision depth perception unit output and dummy head inertial measurement unit output's head inertial data, edge AI analysis module and redundant power and data security module integrate in the protection device shell jointly, the protection device shell is fixed in the automobile body crossbeam.
  2. 2. The system of claim 1, further comprising an inertial wake-up and pose reference unit comprising a high-precision accelerometer for collision triggering, and a high-precision IMU for providing an external reference for pose resolution.
  3. 3. The system of claim 1, wherein the edge AI analysis module comprises a general purpose host processor and a dedicated NPU hardware acceleration unit coupled via a high speed bus for running the pose prediction model.
  4. 4. A bump test dummy head 6D pose data robust acquisition system according to claim 3, wherein said visual depth perception unit comprises a high frame rate camera and a depth camera providing 2D images and 3D depth data.
  5. 5. A collision test dummy head 6D pose data robust acquisition system according to claim 3, wherein said redundant power and data security module comprises a high power density supercapacitor bank, a high speed solid state switch, an impact resistant memory and a data export interface.
  6. 6. A method for robust acquisition of 6D pose data of a head of a crash test dummy, which is applicable to a system for robust acquisition of 6D pose data of a crash test dummy according to any one of claims 1 to 5, and is characterized by comprising the following steps: s1, in a standby and triggering stage, an accelerometer monitors triaxial acceleration of a vehicle continuously with lower power consumption, immediately sends out a hard interrupt signal once a collision precursor is detected, and simultaneously starts a high-frequency data acquisition mode by a dummy head inertia measurement unit; S2, in a visual depth fusion and 2D prediction stage, an edge AI analysis module is started to run YOLOv models; s3, in a real-time 6D gesture resolving stage, the system utilizes YOLOv and PNP algorithm to resolve the 6D gesture of the head of the passenger relative to the vehicle body coordinate system; S4, continuously running an unscented Kalman filtering UKF fusion algorithm in the edge AI analysis module, and taking the 6D gesture, the speed and the acceleration of the dummy head as state variables; S5, in the data processing and analyzing stage, the system generates a 6D gesture sequence of the head of the passenger in real time; And S6, in a data security stage, the power supply is continuously powered, and all high-dimensional 6D gesture data, head 6D gesture and dynamics data after UKF fusion, original sensor data and analysis results are guaranteed to be completely and quickly written into the impact-resistant memory in a very short time after a collision event is ended.
  7. 7. The method for robust collection of 6D pose data of a crash test dummy head according to claim 6, further comprising a post extraction stage, wherein after the crash test, the data can be completely recovered from the memory for analysis through the data export interface and the dedicated recovery tool even if the edge AI module is damaged.
  8. 8. The method for robust acquisition of 6D pose data of a crash test dummy head according to claim 6, wherein the hard interrupt signal in S1 is wake-up edge AI analysis module and instructs a high-speed solid state switch to disconnect a super capacitor from a main power supply and switch to isolated power supply.
  9. 9. The robust acquisition method of the 6D pose data of the head of the crash test dummy according to claim 6, wherein in the step S3, the system utilizes YOLOv to predict 2D key point projection coordinates, combines camera internal parameters and a pre-established 3D model of the head of the human body, and can rapidly calculate the 6D pose of the head of the passenger relative to a coordinate system of the vehicle body through a PnP algorithm by single reasoning.

Description

System and method for robust acquisition of 6D pose data of head of dummy in collision test Technical Field The invention relates to the technical field of data acquisition and analysis of automobile safety collision tests, in particular to a system and a method for robust acquisition of 6D gesture data of a head of a collision test dummy. Background In existing car crash tests, the assessment of occupant protection typically relies on video analysis of sensors and high speed cameras inside the crash dummies. However, the prior art has the disadvantage that visual data is easily interrupted, the deployment of the airbag, and the drastic change in smoke or light rays can severely obscure the high speed camera view at the moment of impact, resulting in loss or blurring of visual data at critical peak moments of the impact (e.g., moment the occupant touches the airbag). The dummy sensor data is limited to fixed points and cannot provide complicated 3D pose, rotation angle, and contour information of the dummy occupant head in a collision. Traditional 2D or 3D keypoint detection has limitations in expressing true translation and rotation (i.e., 6D pose) of the dummy head. Particularly in the critical injury area of the head, the interruption of visual data caused by large-area shielding and the decline of precision caused by drift of the sensor built in the head of the dummy, promote that the continuous and high-precision head kinematics and dynamics data are difficult to acquire under extreme conditions by only relying on the visual or dummy head inertial sensor. The cost and the efficiency are contradictory, the high-precision AI gesture analysis requires larger calculation force and higher cost, and the low-cost general hardware has extremely poor data reliability in collision impact and power-off environments. Especially for 6D gesture calculation, the traditional iterative optimization method is low in speed, and is difficult to meet the real-time requirement of a collision test, so that the system and the method for robust acquisition of the 6D gesture data of the head of the dummy for the collision test are provided. Disclosure of Invention The invention aims to provide a robust acquisition system and method for 6D pose data of a head of a crash test dummy, so as to solve the problems in the background art. In order to solve the technical problems, the invention adopts the following technical scheme: The utility model provides a collision test dummy head 6D gesture data robust acquisition system, includes edge AI analysis module, vision depth perception unit, dummy head inertial measurement unit and redundant power and data security module, the deployment has the unscented Kalman filtering fusion algorithm in the edge AI analysis module for fuse 6D gesture data and the dummy head inertial data of dummy head inertial measurement unit output of vision depth perception unit output, edge AI analysis module and redundant power and data security module integrate in the protection device shell jointly, the protection device shell is fixed in the automobile body crossbeam. Preferably, the system further comprises an inertial wake-up and gesture reference unit, wherein the inertial wake-up and gesture reference unit comprises a high-precision accelerometer for collision triggering and a high-precision IMU for providing external reference for gesture calculation. Preferably, the edge AI analysis module comprises a general main processor and a special NPU hardware acceleration unit connected via a high-speed bus, for running the attitude prediction model. Preferably, the visual depth perception unit comprises a high frame rate camera and a depth camera providing 2D images and 3D depth data. Preferably, the redundant power supply and data protection module comprises a high-power density super capacitor bank, a high-speed solid state switch, an impact-resistant memory and a data export interface. The robust acquisition method of the 6D gesture data of the head of the dummy for the collision test is suitable for a robust acquisition system of the 6D gesture data of the head of the dummy for the collision test, and comprises the following steps: s1, in a standby and triggering stage, an accelerometer monitors triaxial acceleration of a vehicle continuously with lower power consumption, immediately sends out a hard interrupt signal once a collision precursor is detected, and simultaneously starts a high-frequency data acquisition mode by a dummy head inertia measurement unit; S2, in a visual depth fusion and 2D prediction stage, an edge AI analysis module is started to run YOLOv models; s3, in a real-time 6D gesture resolving stage, the system utilizes YOLOv and PNP algorithm to resolve the 6D gesture of the head of the passenger relative to the vehicle body coordinate system; S4, continuously running an unscented Kalman filtering UKF fusion algorithm in the edge AI analysis module, and taking the 6D gesture, the speed