一种基于聚类的深空红外多目标快速检测算法

A Fast Deep-space Infrared Multi-target Detection Algorithm Based on Clustering

该文提出一种基于行扫描点线目标聚类合并的快速实时多目标检测算法。该方法首先对原始图像进行自适应阈值分割,然后采用外接矩补形,点线目标提取和聚类合并对二值图像单帧目标进行全视场检测并编号标记,精度达到像素级,避免了帧差法,投影法等传统检测算法带来的漏检。最后应用五点二次滤波预测目标位置,并构造代价函数进行关联匹配完成目标确认,有效解决了检测中目标分裂,交叉,因重合而暂时消失等问题,提高了系统检测能力。在基于SOPC的硬件平台进行验证,实验结果表明该算法能够准确实时地检测深空目标。

This paper presents a fast real-time multi-target detection algorithm based on line target clustering.Adaptive threshold is applied to image segmentation;And then enclosing rectangle prosthetics,line target extraction and clustering merger are utilized for the binary image to implement full-field pixel-level targets detection and conduct ID tag.So undetected problems caused by the traditional detection algorithm can be avoid;Finally,a five points square predictor and cost function are constructed for trajectory matching,by which the problems of multi-target division,cross,temporarily lost due to overlap and so on are effectively resolved.The experiments are carried on SOPC hardware platform and the results show that the proposed algorithm can perform real-time detection accurately for the deep-space objects.