动态背景下运动车辆的检测与跟踪研究

Research on Detecting and Tracking of Moving Vehicle under Dynamic Background

为实现对车载设备视频图像中车辆的识别和跟踪,针对图像中的运动目标和动态背景,提出了一种基于特征学习的目标检测和超像素跟踪算法。该算法首先对训练图像进行HOG特征提取,并利用AdaBoost算法得到强分类器。利用强分类器对采集的图像进行车辆检测,从而确定搜索区域。结合对搜索区域的超像素分割结果,采用均值漂移聚类算法实现车辆识别与跟踪。实验结果表明,该算法可以很好地实现视频序列中的车辆识别,提高了目标跟踪的实时性。

In order to detect and track moving vehicles in video image of in-vehicle camera, a vehicle detecting and tracking algorithm based on feature learning and super-pixel tracking was proposed in this paper according to moving object and dynamic background in video image. Firstly, HOG feature was extracted from the trained image and AdaBoost algorithm was used to compute reinforced classifier. Then, the reinforced classifier was adopted to detect the captured image to detemaine the search area. Finally, combined with results of super-pixel segmentation, mean-shift clustering algorithm was used to identify and track vehicle. Experimental results show that the proposed algorithm can successfully detect moving vehicle in video image and significantly improve the real-time of vehicle tracking.