基于改进光流法的运动目标检测研究
首发时间:2016-06-29
摘要:运动目标检测在现实场景中具有极其重要的意义,它是跟踪和识别运动物体状态的前提。光流法不需要复杂的背景建模,而且能够得到运动目标位置和运动速度信息,是运动目标检测常用的方法之一。本文针对经典Horn-Schunk光流算法在计算稠密光流时,计算时间长,不能满足实时性要求的问题,提出结合帧差法的改进光流算法,对差值图像中灰度值较大的位置计算光流场分布,其它位置采用迭代平滑。为了验证该算法的有效性,本文将此算法应用于运动目标提取和车流量检测,通过与原始光流法对比来证明本文算法比原始光流算法更精确。其中,在运动目标提取方面,为了准确提取运动目标区域,本文在改进光流法基础上,研究出一种基于连通域分析的方法,得到运动物体最小外包矩形;而车流量检测方面,是对实际拍摄的交通路况视频中的车辆数目进行统计,通过设置虚拟线圈检测其中光流场信息,检测出一段时间内的车流量。
关键词: 光流法 帧差法 运动目标检测 连通域分析; 车流量检测
For information in English, please click here
Research on Moving Target Detection via Improved Optical Flow Method
Abstract:Moving target detection that is the premise of tracking and identification the moving object is extremely important in real-life scenarios. One of the common method used in this field is optical flow method, which does not require complex background modeling, but also is able to get moving target position and velocity information. As classical Horn-Schunk optical flow algorithm spends so much time calculating dense optical flow that can not meet the real-time requirements, this paper proposes an improved optical flow method combined with frame difference, which means the difference image where gray value larger calculate light flow distribution, while other locations use iterative smooth. In this paper, this improved algorithm is applied to moving target detection and vehicle flow detection in order to verify the effectiveness and compare with the original one. Facts have proved that the proposed method is more accurate than the original optical flow algorithm. So as to accurately extract the moving target area, this paper developed a method based on the improved optical flow method with connected component analysis to obtain moving objects minimum bounding rectangle; On the other hand, we count the number of vehicles in the actual shooting of the video traffic conditions by setting the virtual coil and detecting optical flow information, then the statistics for a period of time the traffic is acquired.
Keywords: Optical Flow Method Frame Difference Method Moving Target Detection Connected Domain Analysis Vehicle Flow Detection
论文图表:
引用
No.4698965862381146****
同行评议
共计0人参与
勘误表
基于改进光流法的运动目标检测研究
评论
全部评论0/1000