【摘要】 为了解决多目标跟踪中数据存储量、运算量大与实时处理之间的矛盾,提出了一种基于倒角算子距离变换的多目标分离方法,分别提取出各个目标的位置和大小信息,再采用五点二次预测估计出下一场目标的位置从而得到目标的运动速度,最后由改进的加权最近邻算法进行目标特征关联。在XilinxXCV300E+ADIBlackfin532DSP平台上实现该算法,试验结果表明能在20ms内完成图像处理输出脱靶量,达到实时处理的要求。实验室模拟多目标交叉运动,能准确跟踪原有目标而不受新进入目标的影响。更多还原

【Abstract】 In order to resolve the conflict between excessive data storage、time expensive computation and real-time image processing, a method is proposed based on distance transfer with chamfer to extract the position and scale features of each objects, then a five points square predictor is used to estimate the position of objects next time and compute the velocity. Finally, objects are associated using improved weighed nearest neighbor. The algorithm is implemented on the platform of Xilinx XCV300E +ADI Blackfin 532 DSP. Experimental results show that image processing can be completed within 20 ms and output the miss distance, satisfying the requirement of real-time processing. Tracking crossing objects in lab shows original object can be tracked stably immune from the new entering object.更多还原