基于YOLOv3检测和特征点匹配的多目标跟踪算法  

Multi-target Tracking Algorithm Based on YOLOv3 Detection and Feature Point Matching

作　　者：谭芳 穆平安[1] 马忠雪 TAN Fang;MU Ping-an;MA Zhong-xue(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《计量学报》2021年第2期157-162,共6页Acta Metrologica Sinica

摘　　要：针对传统多目标跟踪算法中行人检测速度慢、易受光照变化、行人快速移动及部分遮挡因素的影响造成行人目标跟踪性能差等问题,提出一种根据经典的Tracking-by-Detection模式,采用深度学习YOLOv3算法检测行人目标,然后利用FAST角点检测算法与BRISK特征点描述算法对相邻帧间的行人目标进行特征点匹配,实现多目标行人跟踪的算法。实验结果表明行人目标在背光、快速移动、部分遮挡等复杂环境下均获得了良好的连续跟踪效果,平均精度达到87.7%,速度达到35帧/s。For multiple target tracking algorithm in the traditional pedestrian detection speed is slow,vulnerable to illumination change,the fast moving of pedestrians and the influence of partial occlusion cause the poor performance of pedestrian target tracking. According to the classic Tracking-by-Detection mode,a new pedestrian tracking algorithm is proposed,which uses deep learning YOLOv3 algorithm to detect pedestrian targets,and then uses fast corner detection algorithm and brisk feature point description algorithm to match the feature points of pedestrian targets between adjacent frames to achieve multi-target pedestrian tracking. The experimental results show that the pedestrian target achieves good continuous tracking effect under various complex environments of backlight,fast movement and partial occlusion,with an average accuracy of 87. 7% and a speed of 35 frames per second.

关 键 词：计量学 多目标跟踪 深度学习 YOLOv3算法 特征点匹配 图像处理 

分 类 号：TB96[机械工程—光学工程]