【摘要】 针对红外多目标检测跟踪中目标遮挡问题和算法实时性与准确性难以兼顾的不足,提出一种快速准确且抗遮挡的红外目标检测跟踪算法。在目标检测阶段,提出一种基于RAS(running Average with selectivity)背景更新的中值滤波背景差分算法。该算法采用中值滤波法建立背景图像,通过引入反馈思想与滑动时间窗模型,使背景更新的实时性与鲁棒性得到改善。同时,为了有效解决目标遮挡难题,提出一种像素投影分离算法,通过对粘连目标的投影曲线进行分析来实现粘连目标的分离。在目标跟踪阶段,通过采用滤波加权均值移位算法,从而有效克服红外目标描述信息不足的缺点。同时,将该算法与Kalman滤波融合,最终实现红外多目标的快速准确跟踪。在不同红外测试集上实验结果表明,所提算法的检测率与正确跟踪率分别提高到91.05%、83.78%,运行速度达到32帧/秒,在抗遮挡性、实时性、准确性与鲁棒性等方面均优于现有的主流算法。更多还原

【Abstract】 An anti-occlusion and real-time algorithm for multi-target detection and tracking applied to infrared images is presented,which is designed not only for meeting the demands of the real-time and accuracy of the algorithm but also for dealing with the failure tracking caused by the occluded targets. In motion detection stage of the targets,a median filtering background subtraction approach based on RAS background updating is presented,which uses the background subtraction approach based on median filter to build the background model and segment the infrared targets. In order to improve the robustness and real-time performance of background updating,the running average with selectivity(RAS) algorithm with the feedback theory and sliding time window model is presented. Meanwhile,in order to effectively solve the target occlusion problem,the method of Pixel Projecting Curve is proposed to separate overlapped targets. In motion tracking of targets,the Mean-Shift algorithm based on weighted filter is adopted to overcome the lack of infrared target feature information. And then the algorithm is fused with Kalman filter to implement fast and accurate tracking of multiple infrared targets. Experiments were conducted on different infrared image test sets. The results indicate that the detection rate and correct tracking rate of the proposed method are up to 91. 05% and 83. 78%,respectively,and the operation speed reaches to 32 frames per second. So the algorithm outperforms the existing mainstream algorithms on anti-occlusion,real-time performance,accuracy and robustness.更多还原