Abstract: The traditional contact measurement of the dynamic tension angle of the seat back in the car whiplash test has problems such as installation limitations and interference with the test response. This paper proposes a non-contact measurement method based on video recognition, using Python+ OpenCV as the technology stack. It adopts the channel and spatial reliability tracker (CSRT) algorithm to track the backrest feature marker points in high-speed video and calculates the angle between the two-point connection vector and the horizontal axis through spatial coordinates. When calculating, first obtain the pixel coordinates of the marked points and complete the planar coordinate conversion. Then, apply inverse trigonometric functions (such as atan2) to solve the angle between them and the horizontal reference, achieving frame-by-frame quantization of the tension angle. The whiplash test results of a certain model of Jiangling show that the Pearson correlation coefficient between this method and the measurement results of Falcon software (whose measurement algorithm adopts the contour fitting method, using the least squares method to fit straight lines and arcs) reaches 0.998 (extremely strong correlation), and the average relative error is 5.27%. Error analysis shows that its accuracy is particularly outstanding in the medium and high tension angle range (>3°) (with an average relative error of <5%), meeting the accuracy requirements for dynamic tension angle measurement of seat backrests stipulated in the China new car assessment programme (C-NCAP) management regulations (2024 Edition) regulation, and it also has the advantages of flexible operation and no contact interference.

Key words: whiplash test; seatback opening angle; video recognition; CSRT algorithm; non-contact measurement

摘要： 汽车鞭打试验中座椅靠背动态张角的传统接触式测量存在安装受限、干扰试验响应的 问题。文章提出一种基于视频识别的非接触式测量方法，以 Python+OpenCV 为技术栈，采用 通道与空间可靠性跟踪器（CSRT）算法跟踪高速视频中靠背特征标记点，通过空间坐标计算 两点连线向量与水平轴的夹角。计算时，首先获取标记点的像素坐标并完成平面坐标转换， 继而应用反三角函数（如 atan2）求解其与水平基准的夹角，实现张角的逐帧量化。江铃某车 型鞭打试验结果表明，该方法与 Falcon 软件（测量算法采用轮廓拟合法，使用最小二乘法拟 合直线和圆弧）测量结果的 Pearson 相关系数达 0.998（属极强相关），平均相对误差为 5.27%。误差分析显示，其在中高张角区间（>3°）的精度尤为突出（平均相对误差<5%）， 满足中国新车评价规程（C-NCAP）2024 版管理规则对座椅靠背动态张角测量的精度要求， 且具备操作灵活、无接触干扰的优势。

关键词: 鞭打试验；座椅靠背张角；视频识别；CSRT 算法；非接触测量