无人机视角下的特征匹配引导粒子滤波跟踪算法  

Features Matching Guided Particle Filter Tracking Algorithm Based on the Drone

作　　者：阚雨婷 施佺 王晗 KAN Yuting;SHI Quan;WANG Han(School of Transportation, Nantong University, Nantong 226019, China;School of Mines, China University of Mining and Technology, Xuzhou 221116, China)

机构地区：[1]南通大学交通学院,江苏南通226019 [2]中国矿业大学矿业工程学院,江苏徐州221116

出　　处：《南通大学学报（自然科学版）》2018年第3期6-11,共6页Journal of Nantong University(Natural Science Edition)　

基　　金：国家自然科学基金项目(61872425);江苏省高校自然科学基金项目(17KJB520029);南通市工业创新项目(GY12016020)

摘　　要：粒子滤波(particle filter, PF)算法被广泛应用于视觉目标的跟踪,然而,在无人机视角下,摄像机与画面中的目标同时运动,导致了PF对目标运动状态的预测失效.针对此问题,提出一种面向无人机视角下的改进的粒子滤波跟踪算法——特征匹配引导的粒子滤波跟踪算法.首先,利用相邻两帧图像中目标物体尺度不变特征变换(scale invariant feature transform, SIFT)特征匹配的结果作为初次定位;然后,利用空间加权的HOG特征与PF相结合获取二次定位结果;最后,利用chamfer distance修正跟踪结果的SIFT特征点作为下一帧特征匹配的模板,从而循环产生准确的视频跟踪结果.比较试验表明,该算法有效地改善了传统PF跟踪算法在无人机视角下运动状态预测方程失效的问题,能够较准确地对运动目标进行跟踪.Particle filter(PF) has been widely applied to visual object tracking. However, due to the simultaneous movement of the camera and the detected object, the traditional PF tracking algorithm for the fixed camera is disabled. Aiming at this problem, this paper proposes an improved PF algorithm: features matching guided particle filtering algorithm(FMG-PF), which can effectively track the targets. Firstly, scale invariant feature transform(SIFT)features matching result is used as initial localization. Then it is corrected by combination of spatial weighted HOG feature and particle filter frame as tracking result. Finally, chamfer distance is employed to modify the feature points of tracked target and the revised target feature points are used as the features template for the next frame matching.Experiments show that the proposed algorithm is able to track the road traffic targets effectively.

关 键 词：特征匹配引导 粒子滤波 无人机 目标跟踪 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]