The accuracy of accident reconstruction depends on the integrity and details of measurement of the accident scene. The photographs of accident scene and vehicles contain details of accident scene. Two-eyes vision in computer can be used to obtain the information from these photographs. This paper introduces the principle of two-eyes vision in computer, which can be used in measuring accident scene from photographs. Because of maneuverability of investigating accident scenes, a special reference object had been used to calibrate the camera parameters. The errors of image matching affect the precision of reconstruction greatly, so image matching was taken based on correlation of two projective points in two photographs of the same space point in the accident scene. The full-scale models of traffic accident scenes had been measured using this method, and it proved that the traffic accident scenes can been measured using this method.

A new method for line detection based on chain code detection is proposed. First of all, chain codes are tracked in an edge image. After the curvature estimated, the corners in chain codes can be detected, and chain codes are separated at the corners. Chain codes of line segments are recognized, and only the chain codes, which represent line segments, are kept. Parameters of line segments are estimated according to the chain codes, and the piecewise line segments are linked based on the linking criterions. Experiments with images show that the method is effective for line detection.

变形车辆测量是道路交通事故再现的重要基础性工作。建立基于普通数码相机的事故变形车辆自标定三维重建模型，可以避免在现场设置标定物，有利于提高变形车辆的勘测效率和三维重建设备的使用方便性。研究了基于基础矩阵估计的事故变形车辆摄影图像自标定三维模型，并分析了其实现过程。实验表明，自标定三维重建能为事故再现提供详细的车辆变形描述。

论述了基于普通数码相机的交通事故摄影测量技术的基本理论，提出了相应的关键技术并对其研究现状进行了总结分析。根据交通事故再现分析的需要和基于普通数码相机的事故摄影图像特点，提出了进一步研究的内容和主要方法。

为避免测量过程中在现场设置标定参考点，研究了以基础矩阵估计为基础的交通事故摄影图像自标定模型，并分析了其实现过程。把便于递推实现的无限对称指数滤波器引入平均梯度矩阵计算，实现了图像特征点的检测；通过匹配强度系数计算，对两幅图像的特征点进行了匹配；分析了基础矩阵计算及不良匹配点剔除的最小中值二乘法；对通过本质矩阵分解标定摄影图像的实现过程进行了论述。通过实验对模型及其实现方法进行了验证。