运动目标最优角点选择算法

A New Algorithm for Selecting Optimal Corner Point of Moving Target

针对传统角点提取算法应用于运动目标优质角点选取上,较难平衡检测精度和检测速率的问题,提出一种稳健的角点选择新算法。首先将图像分块,用子块在相邻帧间的灰度差对其进行聚类,并分割出运动区域块作为角点的搜索区域;然后引入邻域块重新构造Moravec算法的能量变化计算方法,并用之检测目标上的角点;最后对检测到的所有角点进行动态过滤,剔除掉质量低的角点,留下质量高的角点代表运动目标的局部显著特征。将本算法应用于目标跟踪系统进行测试,测试结果表明它有较好的稳健性,较强的的抗噪能力和较快的检测速度,且能保证跟踪的准确性,可以很好地满足交通场景中对行人检测实时性和可靠性的要求。

Using traditional corner extraction algorithm on optimal motion target corner detection,it is difficult to balance the detection accuracy and detection speed,in order to solve this problem,a novel and robust algorithm based on motion region segmentation was proposed. Firstly,the algorithm divided image into several blocks,each block employed its gray difference in adjacent frames to judge whether it was a motion region block or not,if it was the motion region block,split it out as a search area of corner. Then,with the concept of block,it changed Moravec algorithm to reconstruct the energy calculation method,and used it to detect the corner. Finally,delimited the influence scope of the optimal corners,and through dynamic filtering algorithm eliminated the excess corners in range of them. The results in tracking system show that the proposed algorithm has great robustness and noise suppression capability,and the corners extracted can assure accurate tracking.