基于遮挡重检测的OCC系统目标LED阵列跟踪算法  

Investigation of target LED array tracking algorithm in the optical camera communication system based on occlusion redetection

作　　者：闫晓明 韩倩倩 籍风磊[1] YAN Xiaoming;HAN Qianqian;JI Fengei(School of Communication Engineering,Jilin University,Changchun 130012,China;Tianjin Aviation Electromechanical Co.,Ltd.,Tianjin 300308,China)

机构地区：[1]吉林大学通信工程学院,吉林长春130012 [2]天津航空机电有限公司,天津300308

出　　处：《实验技术与管理》2024年第11期8-15,共8页Experimental Technology and Management

基　　金：国家自然科学基金项目(62271228);吉林省科技发展计划重点研发项目(20200401122GX);吉林大学实验技术重点项目(SYXM2024a009)。

摘　　要：在光学相机通信(OCC)系统对目标发光二极管(LED)阵列跟踪过程中,复杂的室外环境会严重影响其性能;同时,由于目标LED阵列持续不断发送信息,即每一帧目标LED阵列明暗闪烁状态都有可能不同,这对目标LED陈列跟踪算法的研究带来了极大的挑战。论文针对此问题提出了一种基于遮挡重检测的跟踪算法。该算法在核相关滤波算法的基础上采用2种特征融合互补和加入尺度滤波器的方式提升算法的抗干扰能力,同时结合目标遮挡判别机制和改进轻量化Yolov5目标检测算法提升算法的抗遮挡能力。实验结果表明,该算法在室外光照变化、尺度变化和遮挡场景下,平均中心位置误差最大为8.94个像素,跟踪成功率最低为92.4%,平均帧率最低为98.18帧/s。[Objective]During the tracking process of a target LED array,the complex outdoor environment will seriously affect the optical camera communication(OCC)system’s performance,for example,light changes,scale changes,and target occlusion scenes may lead to poor tracking effects,tracking drift,or even tracking failure.Furthermore,because the target LED array continuously sends information,its flickering state may be different in each frame,which poses a great challenge to the research of target tracking algorithms.In this study,an object-tracking algorithm that incorporates occlusion awareness and re-detection is proposed to address this problem.[Methods]Based on the kernel correlation filter,this study proposes feature fusion and scale adaptation for the object-tracking algorithm.In this algorithm,the fast histogram of the oriented gradient feature instead of the histogram of the oriented gradient feature is used to reduce computational complexity,additionally,the rotate local binary pattern feature is integrated to improve the algorithm’s ability to express target information.A scale filter that improves the adaptability of the algorithm to scale changes is added to judge the target scale.The addition of two types of feature fusion and complementary and the scale filter improves the anti-jamming ability of the algorithm.To tackle the poor anti-obscuration ability of feature fusion and scale adaptation for object-tracking algorithms,an occlusion-aware and re-detection object-tracking algorithm is proposed.In the algorithm,a target occlusion discrimination mechanism is implemented to determine whether a target is occluded.When occlusion occurs,a target detection algorithm is employed to detect the target LED array.Considering the limited hardware resources of the OCC system platform,three lightweight networks—MobileNetV3,ShuffleNetV2,and GhostNet—are utilized to replace the backbone network in the Yolov5 network structure.By comparing and analyzing the detection accuracy,real-time performance,and network compl

关 键 词：光学相机通信 目标跟踪 特征融合 多尺度 遮挡判断 目标重检测 

分 类 号：TN929.12[电子电信—通信与信息系统]