基于无人机航拍视频车辆多目标跟踪算法研究  

Research on Vehicle Multi-target Tracking Algorithm Based on UAV Aerial Video

作　　者：朱奇光[1,2] 商健 刘博 岑强[1] 陈卫东 ZHU Qiguang;SHANG Jian;LIU Bo;CEN Qiang;CHEN Weidong(School of Information Science and Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China;Hebei Provincial Key Laboratory of Special Optical Fibers and Fiber Optic Sensing,Qinhuangdao,Hebei 066004,China)

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004 [2]河北省特种光纤与光纤传感重点实验室,河北秦皇岛066004

出　　处：《计量学报》2024年第12期1772-1779,共8页Acta Metrologica Sinica

基　　金：国家自然科学基金(61773333,62273296)。

摘　　要：为了提高无人机视觉平台下的车辆多目标跟踪精度,提出了一种改进YOLOv7网络与优化ByteTrack算法相结合的无人机视觉车辆多目标跟踪算法。首先,针对小目标特征不明显的情况,增强了YOLOv7网络浅层语义信息的特征提取能力,同时采用SIoU-Loss对坐标损失函数进行优化,加快锚框收敛速度;其次,根据车辆运动特点,在ByteTrack算法的基础上,将卡尔曼滤波算法的状态向量融入加速度信息;最后,在VisDrone2021数据集上验证算法的有效性。实验结果表明:改进YOLOv7网络的平均检测精度比原网络提高3.2%,跟踪算法准确度比基准算法提高1.2%,高阶跟踪精度提高2.9%。In order to improve the vehicle multi-target tracking accuracy under the UAV vision platform,a UAV visual vehicle multi-target tracking algorithm that combines the improved YOLOv7 network with the optimized ByteTrack algorithm is proposed.Firstly,in view of the situation where the features of small targets are not obvious,the feature extraction ability of shallow semantic information of the YOLOv7 network is enhanced,and SIoU-Loss is used to optimize the coordinate loss function to speed up the convergence speed of the anchor frame.secondly,according to the vehicle motion characteristics,in based on the ByteTrack algorithm,the state vector of the Kalman filter algorithm is integrated into the acceleration information.finally,the effectiveness of the algorithm is verified on the VisDrone2021 data set.The experimental results indicate that the average detection accuracy of the improved YOLOv7 network is 3.2%higher than the original network,the accuracy of the tracking algorithm is 1.2%higher than the baseline algorithm,and the high-order tracking accuracy is improved by 2.9%.

关 键 词：计算机视觉 图像处理 多目标跟踪 无人机 YOLOv7网络 ByteTrack算法 车辆检测 

分 类 号：TB96[机械工程—光学工程]