基于时空坐标图像的运动点目标检测方法  

Moving point target detection based on spatiotemporal image

作　　者：廖媛[1] 孙胜利[1] 陈桂林[1] 

机构地区：[1]中国科学院上海技术物理研究所

出　　处：《光电工程》2007年第1期23-26,36,共5页Opto-Electronic Engineering

摘　　要：针对复杂背景下运动点目标检测方法存在的局限性,提出从时间空间角度检测运动目标的方法。在时间空间域里,背景的复杂纹理表现为无限长直线,运动目标则表现为有限长线段。针对该特点,运用逐点灰度均值将背景与运动点目标分离,并根据投影关系将目标投影到二维空间,获得直观的目标运动轨迹。试验结果表明,该方法能较好地从复杂背景中分离出速度不大于1像素/帧的点目标,且可同时进行多目标的检测,抗噪声能力强。In view of the limitation existing in detecting moving point targets in complex background, a new detection method with detailed theory based on spatiotemporal image is presented. In 2D spatiotemporal image, moving object shows finite line segment and complicated texture of background displays infinite straight line. In accordance with this characteristics, moving objects can be effectively extracted with local auto-adaptive threshold, and the trajectory of moving object can be gotten by applying projective transformation from spatiotemporal space to geography space. Experimental results show that the method can effectively detect low SNR point targets in complex background at the speed of less than lpixel/frame. Furthermore, multi-target detection can be performed simultaneously.

关 键 词：运动目标 目标检测 点目标 序列图像 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]