基于特征光流的多运动目标检测跟踪算法与评价

光流车辆检测算法其光流不仅携带了运动物体的运动信息,还包含丰富的三维结构信息,能够在未知场景信息的情况下对运动目标进行准确检测;但传统光流法计算方法复杂、抗噪性能差、处理速度缓慢,无法满足多目标实时检测的实际需求。为提高光流法实时检测效率,同时保持较好的检测精度,提出了一种基于Harris特征点光流及卡尔曼滤波模型的多运动目标跟踪算法;并提出新的视频目标检测算法性能评价指标。通过对不同实验场景下多个运动目标的检测与跟踪实验统计结果表明,对比主流Meanshift车辆跟踪算法,检测精度平均提高4.61%;且跟踪持续性提升41.5%,具有更好的鲁棒性及准确性。在时间效率上较比传统光流法平均提升42.9%,能够更好地满足目标跟踪实时性要求。

Multi-objective Detection and Tracking Algorithm Based on Feature Optical Flow

The optical flow algorithm of detecting vehicles not only carries the motion information of moving objects but also contains abundant information of three-dimensional structure so that it could accurately detect moving targets in the case of not knowing the scenario or information. However, the traditional optical flow algorithm is complicated, poor in anti-noise performance and slow in processing speed, so it could not meet the actual demands of detecting multiple targets in real time. In order to improve the real-time detection efficiency of optical flow algorithm while maintaining the good accuracy of detection, this thesis proposed an algorithm of tracking multiple moving targets based on Harris feature optical flow and Kalman filtering model and also put forward new performance evaluation indexes for the algorithm of detecting video targets. The statistical results of detecting and tracking multiple moving targets in different experimental scenarios show that: compared with the mainstream algorithm of tracking Meanshift vehicles, the detection accuracy of this algorithm increases by 4.61% on average and its persistence of tracking increases by 41.5% with a better robustness and accuracy. In terms of time efficiency, it increases by 42.9% on average compared with the traditional optical flow algorithm, so it could better meet the real-time demand of tracking targets.