论文部分内容阅读
为提高车辆跟踪的准确性,满足智能交通监控的需求,采用样本车型模板结合尺度不变特征变换(SIFT)算法匹配检测车辆位置,并将模板大小与车辆尺度变化联系起来,根据特征向量距离最终获得车辆准确区域;采用结合均值漂移的粒子滤波算法对车辆进行跟踪,根据跟踪尺度改变跟踪窗口大小,并通过独立粒子滤波建立了多运动车辆之间的数据关联。实际道路测试结果表明:该算法的车辆检测准确率达到90%以上,特征粒子在后续跟踪过程中状态稳定,漂移在跟踪窗口外的粒子数都保持在10%以下。
In order to improve the accuracy of vehicle tracking and meet the demand of intelligent traffic monitoring, the SIFT algorithm was used to match and detect the position of the vehicle, and the template size was correlated with the vehicle scale change. According to the final distance of eigenvectors, The vehicle’s exact region is obtained. The particle filter algorithm based on mean shift is used to track the vehicle, and the tracking window size is changed according to the tracking scale. The data association among multi-sport vehicles is established through independent particle filter. The actual road test results show that the proposed algorithm can detect vehicle more than 90% accurately, and the characteristic particles keep steady state in the follow-up process. The number of particles drifting outside the tracking window keeps below 10%.