一种运动目标检测实时性和鲁棒性的方法

摘要

In the paper we propose an improved algorithm for moving objects detection based on both the two-frame difference method and the background subtraction method targeting at the deficiencies of them respectively.First,we make use of two-frame difference method to carry out coarse segmentation on moving regions,and execute morphological dilation processing to eliminate the influence of "hollow".Then we present the improved adaptive background updating method,which is based on movement area ratio.Finally,we employ background frame subtraction method to make refined segmentation on motion region segmented in coarse segmentation.This method retains the real-time per-formance characteristics of two-frame difference method and background frame subtraction method,meanwhile it also solves the "hollow","double shadow"phenomenon in two-frame difference method as well as the misjudgement of background updating phenomena in background frame subtraction method caused by light changing,thus has strong robustness.%针对二帧差法和背景差法各自的不足提出建立在这两种方法基础之上的改进算法。该方法首先使用二帧差法对运动区域进行粗分割，并进行形态学膨胀处理来消除“空洞”影响。然后提出基于运动面积比的改进的自适应背景更新方法。最后采用背景帧差法对粗分割出来的运动区域进行细分割。该方法既保留了二帧差法和背景帧差法实时性能特点，同时解决了二帧差法中产生的“空洞”、“双影”现象和背景帧差法中由于光照变化等造成的背景更新误判现象，具有很强的鲁棒性。