光照强度突变场景下的三维目标跟踪研究  

Research on 3D Target Tracking in the Scene of Sudden Change of Illumination Intensity

作　　者：李庆楠 丁智[1] LI Qingnan;DING Zhi(School of Art and Design,Wuhan University of Technology,Wuhan 430070,China;National Engineering Research Center for Multimedia Software,School of Computer Science,Wuhan University,Wuhan 430072,China)

机构地区：[1]武汉理工大学艺术与设计学院,武汉430070 [2]武汉大学国家多媒体软件工程技术研究中心,武汉430072

出　　处：《武汉理工大学学报（交通科学与工程版）》2023年第1期42-46,共5页Journal of Wuhan University of Technology（Transportation Science & Engineering）

基　　金：国家自然科学基金(61502348,61671336,91738302)。

摘　　要：文中基于多源特征的卷积神经网络,通过挖掘智能体位置、速度、方向协同变化中的运动行为知识,设计基于融合特征信息熵的目标跟踪决策模型,校准三维目标跟踪准确度.在nuScenes自动驾驶公开数据集、路桥隧道数据集上进行验证,结果表明:该算法相比现有方法,在光照强度突变场景下关键指标AMOTA显著提升.Based on the convolutional neural network with multi-source features, a target tracking decision-making model based on the fusion of feature information entropy was designed by mining the knowledge of motion behavior in the coordinated change of position, speed and direction of agents, and the accuracy of three-dimensional target tracking was calibrated. It is verified on nuScenes autopilot public data set and road and bridge tunnel data set. The results show that compared with the existing methods, the key index AMOTA is significantly improved in the scene of sudden change of light intensity.

关 键 词：三维目标跟踪 行为知识 信息熵 光照强度突变 卷积神经网络 

分 类 号：U664[交通运输工程—船舶及航道工程]