视频中多目标车辆的检测与跟踪方法研究

Research on Method of Detection and Tracking of Multi-objective Vehicle Based on Video

为了提高行驶车辆检测与追踪的准确性、时效性,提出了一个基于熵值法求权重的Adaboost和帧差法的混合检测模型。首先对路面车道线进行识别与分割,将当前摄像头所在的路面作为感兴趣区域提取出来;然后对区域内的车辆使用该模型进行检测与定位;最后使用改进的KCF算法对检测到的车辆进行追踪。实验结果证明,该方法可有效地滤除其他非目标物体,对车辆进行有效地筛选,准确率达95.5%,且检测效率较高,适合在路面场景下对汽车进行有效的检测与追踪。该模型优化了计算资源的使用数量,并且保证了在大量的图像帧中快速地识别与追踪车辆。

In order to improve the accuracy and timeliness of the moving vehicles＇ detection and tracking,we propose a hybrid detection model based on Adaboost of obtaining weight by entropy method and frame difference. Firstly,we identify and segment the road lanes,taking the road where the camera is located in as the region of interest.Then the model is used to detect and locate the vehicles of the region.Finally the detected vehicles are tracked by the improved KCF algorithm.Experiment shows that this method can effectively filter out other non-target objects and screen the vehicles,and its accuracy can reach by 95.5 percent,which has high measuring efficiency and can be used to detect and track vehicles effectively in the road scene.It optimizes the usage quantity of computing resource,and guarantees the quick recognition and vehicle tracking in a large number of image frames.