基于光流的动态背景运动目标检测算法

Detection of Moving Target in Dynamic Background Based on Optical Flow

Horn-Schunch(HS)光流算法能够在动态背景中检测出完整的运动目标,但该算法计算量大,且在图像纹理平滑区域不能检测出光流。本文在分析HS算法运算量的基础上,提出一种结合金字塔Lucas-Kanade(LK)光流和HS光流的动态场景运动目标检测算法。该算法首先利用金字塔LK光流法计算出图像的稀疏光流,根据稀疏光流的运动方向和幅值大小去除运动目标和误匹配点的运动矢量,提取出背景的运动矢量作为HS光流计算的初始值,完整地检测出运动目标。实验结果表明,该算法有效地提高了运动目标检测的速度,适用于云台摄像机的智能监控。

Horn-Schunch（HS） optical flow algorithm can detect the whole moving object in the dynamic background.However,the complexity of the algorithm is great and it can not detect the optical flow in the region where the texture is smooth in the image.Based on the analysis on HS algorithm,this paper proposed an algorithm combined with the pyramid Lucas-Kanade（pyrLK） optical flow and the HS optical flow to detect the moving object in dynamic scene.In this paper,we firstly calculate the sparse optical flow of the image using the pyrLK optical flow algorithm.Then we eliminate the moving vector of the moving object and the mismatched points in terms of the moving direction and the amplitude of the sparse optical flow.At last,we take the moving vector extracted from background as the initial value for HS optical flow calculation to detect the whole moving object.The experiment result shows that our algorithm improved the speed of the detection for moving object and it fits the smart surveillance with the Pan/Tilt camera.