Abstract: Objective In order to solve the problem of feature reference selection and placement caused by changes in the position of surveillance cameras in video speed identification, a feature reference layout method for road traffic accident reconstruction is proposed using image homography. Methods Firstly, we examined the road surveillance video cameras and roughly determined the location of feature placement. Secondly, we set up feature references on site and use drones/cameras to obtain on-site photos containing feature references. Thirdly, we solved the homography matrix and converted on-site photos into surveillance visual angles. Finally, we completed the fusion and matching of on-site photos and surveillance videos. Results The deployment of feature reference in surveillance videos was realized, and the effectiveness of the video feature reference layout for road traffic accident reconstruction was verified. Conclusion This method can be used in the forensic identification of vehicle driving speed and determining the driving boundary with position changed surveillance cameras, and can effectively serve the road traffic accident investigation and identification work using video images.

Key words: road traffic accident, vehicle speed identification, surveillance video, viewpoint transformation, feature 
reference

摘要： 目的 为克服视频车速鉴定中由于监控相机位置变动造成的特征参照难以有效选取、布设的问题，利用图像单应性关系，提出一种面向道路交通事故重建的视频特征参照布设方法。方法 首先，检视道路监控视频，大体确定特征参照布设位置；其次，现场布设特征参照，并利用无人机/相机获取包含特征参照的现场照片；再次，求解单应性矩阵，将现场照片转换为监控视频视角；最后，完成现场照片与监控视频的融合匹配。结果 通过在道路交通事故鉴定实例中的应用，实现特征参照在监控视频中的布设，验证了面向道路交通事故重建的视频特征参照布设方法的有效性。结论 该方法可满足监控相机位置变动情况下车辆行驶速度的求解及行驶边界的确定等需求，可有效服务于利用视频图像的道路交通事故司法鉴定工作。

关键词: 道路交通事故, 车速鉴定, 监控视频, 视角变换, 特征参照