联合YOLOV3和核相关滤波算法的红外视频目标检测及跟踪  

INFRARED VIDEO OBJECT DETECTION AND TRACKING COMBINED YOLOV3 WITH KERNEL CORRELATION FILTERING ALGORITHM

作　　者：戴亚峰 陶青川[1] 杨波[1] Dai Yafeng;Tao Qingchuan;Yang Bo(College of Electronics and Information Engineering,Sichuan University,Chengdu 610065,Sichuan,China)

机构地区：[1]四川大学电子信息学院,四川成都610065

出　　处：《计算机应用与软件》2022年第1期200-205,230,共7页Computer Applications and Software

摘　　要：红外热成像视频与可见光视频相比噪点较多,细节模糊,缺少颜色特征,传统算法常出现目标检测错误、目标跟踪丢失的情况。对此,将YOLOV3目标检测算法与核相关滤波算法相结合,进行红外视频目标的检测及跟踪任务。加载训练后的YOLOV3网络模型进行第一帧目标检测,完成目标跟踪的初始化目标选取,使用KCF算法对后续视频帧进行目标跟踪。在跟踪目标丢失时启动目标再检测,匹配丢失目标并调整跟踪器。实验结果表明,该方法能够克服目标跟踪丢失的情况,且在实验平台上计算速度可达到实时要求。Compared with visible light video,infrared video has more noise and blurred details,and lacks color features.Traditional algorithms often suffer from target detection errors and target tracking loss.We combine YOLOV3 target detection algorithm with kernel correlation filtering algorithm to perform infrared video target detection and tracking tasks.The trained YOLOV3 network model was loaded for the first frame target detection.The model completed the initial target selection for target tracking,and used KCF algorithm for target tracking for subsequent video frames.When the tracking target was lost,the model started target re-detection,matched the lost target and adjusted the tracker.The experimental results show that the proposed method can overcome the loss of target tracking,and the calculation speed can meet the real-time requirements on the experimental platform.

关 键 词：目标跟踪 红外视频 核相关滤波 智能监控 目标检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]