基于稠密光流算法的运动目标检测的Python实现  

DOF algorithm based moving object detection programmed by Python

作　　者：欧阳玉梅[1] OUYANG Yumei(Institute of Information Technology of GUET,Guilin 541004,China)

机构地区：[1]桂林电子科技大学信息科技学院,广西桂林541004

出　　处：《现代电子技术》2021年第1期78-82,共5页Modern Electronics Technique

基　　金：2018年度广西高校中青年教师基础能力提升项目(2018KY0828)。

摘　　要：在运动目标识别算法中,帧间差分法、背景差分法容易出现“重影”“空洞”及“拖尾”现象从而导致识别准确率低。将稠密光流法应用于运动目标检测,基于PyCharm开发环境建立了一个运动目标检测系统,研究采用Gunner Farneback稠密光流法(DOF)计算各像素点位移矢量的光流矩阵,将光流信息转化到HSV空间,并利用Sobel算子进行边缘检测以提高检测效果。实验结果表明,在运动目标与背景对比度低的场景中及速率不同的多运动目标场景中,所提算法皆能实现准确识别与追踪,而且避免了“重影”“空洞”及“拖尾”问题,具有很好的抗干扰能力和识别精确性。In the algorithms of moving target detection,the inter⁃frame difference method and the background difference method are prone to the occurrence of"ghosting","hollow"and"trailing",so the recognition accuracy is low.A moving target detection system is built on the basis of PyCharm development environment and by utilizing dense optical flow(DOF)algorithm.In the study,Gunner Farneback DOF method is used to calculate the optical flow matrix of displacement vector of each pixel point,the optical flow information is converted into HSV space,and then the Sobel operator is used for edge detection to improve the detection effect.The experimental results show this algorithm can achieve accurate recognition and tracking in scenes with low contrast between moving target and background,and in scenes with multiple moving targets at different speeds,avoid"ghosting","hollow"and"trailing"phenomena,and has good anti⁃interference ability and high recognition accuracy.

关 键 词：运动目标检测 稠密光流算法 光流矩阵计算 光流信息转化 边缘检测 系统设计 

分 类 号：TN911.1-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]