基于精确背景运动补偿的机器人运动目标检测与跟踪  

Robot Moving Target Detection and Tracking Based on Accurate Background Motion Compensation

作　　者：马东[1,2] 刘超杰 王立玲[1,2] 王洪瑞[1] MA Dong;LIU Chaojie;WANG Liling;WANG Hongrui(College of Electronic and Information Engineering,Hebei University,Baoding Hebei 071002,China;Robotics Research Center,Hebei University,Baoding Hebei 071002,China)

机构地区：[1]河北大学电子信息工程学院,河北保定071002 [2]河北大学机器人技术研究中心,河北保定071002

出　　处：《机床与液压》2023年第3期20-27,共8页Machine Tool & Hydraulics

基　　金：国家自然科学基金青年科学基金项目(61703133);国家重点研发计划(2017YFB1401200)。

摘　　要：为实现动态场景下机器人快速准确地检测与跟踪运动目标,提出一种基于精确背景补偿的运动目标检测方法,并利用Kalman滤波扩展的KCF算法进行目标跟踪。针对传统ORB算法存在特征点分布不均匀、误匹配率高导致背景补偿效果不佳的问题,采用小波变换及图像分块处理保证提取的特征点数目及均匀分布,通过SURF算法提取具有尺度不变性的特征点并构建ORB描述子。利用KNN算法与对称约束相结合的特征匹配法提高匹配精度,同时引入改进的RANSAC方法精确求解全局运动参数完成背景运动补偿,最后通过帧差法及形态学处理获得完整的运动目标,并将KCF算法融合Kalman滤波实现对被遮挡目标进行再跟踪。实验结果表明:该方法特征匹配正确率达到98.4%,在背景运动状态下能够实时准确地检测出运动目标,并且能够连续稳定地跟踪目标。In order to realize the fast and accurate detection and tracking of moving target by robot in dynamic scene, a moving target detection method based on accurate background compensation was proposed, and the extended Kalman filter KCF algorithm was used for target tracking.In view of the poor background compensation effect caused by uneven distribution of feature points and high false matching rate in the traditional ORB algorithm, wavelet transform and image block processing were used to ensure the number and uniform distribution of extracted feature points.The scale invariant feature points were extracted by SURF algorithm and the ORB descriptor was constructed.The feature matching method that combined KNN algorithm and symmetry constraint was used to improve the matching accuracy.At the same time, the improved RANSAC method was introduced to accurately solve the global motion parameters to complete the background motion compensation.Finally, the complete moving target was obtained by frame difference method and morphological processing, and the KCF algorithm combined with Kalman filter was used to realize the retracking of the occluded target.The experimental results show that the feature matching accuracy of the method is 98.4%,the moving target can be detected in real time and accurately in the background moving state, and the target can be tracked continuously and stably.

关 键 词：动态场景 背景补偿 运动目标检测 目标跟踪 ORB算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]