基于机器视觉车祸车辆的识别方法研究

Based on Machine Vision Accident Vehicle Recognition Method Research

在交通安全中对车祸车辆进行智能优化识别具有重要意义。由于采集的车祸车辆信息容易受到光照及速度的影响,使得不能准确收集车辆特征。传统的识别方法,主要根据二维图像的车体轮廓特征进行识别,当车辆轮廓发生动态化时,不能准确的对车祸中车辆轮廓特征进行收集和预处理,导致识别精确度低的问题。提出嵌入式机器视觉车祸车辆的识别方法,对车祸车辆图像进行几何模型的构建,获得交通事故车体轮廓图像特征。再将二维平面图像进行转换,引入嵌入式机器视觉对车体外轮廓变化点进行定位得到车体动态特征构成点及受撞击的车体轮廓图像区域。在采用欧式距离度量,计算车体图像集序列均值之间的长度。通过最近邻分类器(NN)的边缘检测,确定车祸中车体轮廓变化位置,实现车祸车辆的识别。通过仿真可知,改进方法能够准确的对相应的车祸车体轮廓进行识别,相比传统识别方法识别精度高。

An identification method of the accident vehicle was proposed based on embedded machine vision. A geometric model of accident vehicle image was constructed to obtain the body contour image features of the traffic accident vehicle. Then two-dimensional image was transformed. Embedded machine vision was introduced to position the change point of external body contour, and obtain the constitute point of dynamic characteristics of a body and the im- pacted body contour image area. By using the Euclidean distance metric, the length between the mean of body＇ s im- age sequences of sets can be computed. Through edge detection of the nearest neighbor classifier （NN）, the changed position of body contour of accident vehicle was determined, and the accident vehicle identification was achieved. Simulation results show that the improved method can accurately identify the corresponding body contour of accident vehicle, and has high accuracy compared with the traditional identification method.