基于多线索的车辆跟踪方法研究

针对车载摄像机的汽车辅助驾驶系统,提出了一种基于多线索的车辆跟踪算法。利用运动估计预测目标在下一帧的感兴趣区域,即跟踪窗。在跟踪窗内联合使用车辆的先验知识、评估环节、分块模板等线索定位车辆目标。在跟踪过程中,提出去除误识别的措施。这种基于多线索的车辆跟踪方法实现了对后方多车辆的稳定跟踪,也适合前视车辆。实验结果表明该算法是鲁棒和准确的。

Research on Vehicle Tracking Algorithm Based on Multiple Clues

The multi-clue vehicle tracking algorithm in image sequences was proposed for driver assistance system. Images were acquired by a camera installed at a moving vehicle. Firstly the regions of interest which contained the target in the next frame, namely tracking windows, were predicted by motion estimator. Multiple clues, such as prior knowledge of vehicle, assessment component, division block template etc, were integrated to locate the vehicle target in the tracking window. Measure of removing false positive was proposed in the tracking process. The vehicle observed in rear view could be tracked stably by the multi-clue vehicle tracking algorithm. The approach is suitable for tracking vehicles observed in frontal view. Experiment results illustrate the robust and accurate performance of the tracking algorithm.