融合视觉里程计和BP神经网络的自适应行人航迹推算方法  

VPO:Visual Odometry-assisted Pedestrian Dead Reckoning Based on Neural Network and Extended Kalman Filter

作　　者：李晋 陈威 刘羽鹤 高瑞雪 冯立辉[2] Li Jin;Chen Wei;Liu Yuhe;Gao Ruixue;Feng lihui(School of Beijing,Beijing Institute of Technology,Beijing,102401,China;School of Optics and Photonics,Beijing Institute of Technology,Beijing,100081,China;School of Integrated Circuits and Electronics,Beijing Institute of Technology,Beijing,100081,China)

机构地区：[1]北京理工大学北京学院,北京102401 [2]北京理工大学光电学院信息光子技术工信部重点实验室,北京100081 [3]北京理工大学集成电路与电子学院,北京100081

出　　处：《中国现代教育装备》2024年第7期27-32,共6页China Modern Educational Equipment

摘　　要：当前行人导航定位技术被越来越广泛应用于民用XR领域和军用单兵作战系统,其对定位追踪精度和复杂环境下系统的鲁棒性提出了更高要求。提出了一种融合视觉里程计(Visual Odometry,VO)、反向传播神经网络(Back Propagation Neural Network,BP)和行人航迹推算(Pedestrian Dead Reckoning,PDR)即联合视觉的行人航迹推算里程计(Visual PDR Odometry,VPO),并基于此进行扩展卡尔曼滤波。利用胸口装备惯性传感器和RGB-D深度相机采集行进数据,输入VO和PDR模块得出各自的运动位姿。将有效的步长、航向角、加速度幅值、角速度平均值及步频作为BP神经网络的训练数据集进行网络训练。当VO失效时,BP计算数据可作为备选观测值,将其与PDR数据一起导入扩展卡尔曼滤波器中进行数据融合。实验结果表明,行人航迹误差为1.026‰,优于经典VO和PDR推算方法,能在复杂环境下提高系统的鲁棒性。Currently,as the pedestrian navigation and positioning technology is more widely used in the field of XR and individual combat system,users put forward higher requirements for its tracking accuracy and robustness in complex environment.This paper proposes an algorithm framework,VPO,based on extended Kalman filter,which combines the tracking results from visual odometry(VO),back propagation neural network(BP)and pedestrian dead reckoning(PDR).VPO obtains the tracking data of VO and PDR respectively from the tracker with RGB-D camera and IMU which is equipped on soldier's chest.Then,the qualified VO results will be used as training set of BP together with acceleration,angular velocity and step frequency.In case of VO lost,BP results can be used as observation results,and be imported into extended Kalman filter together with PDR data for fusion.The experiment results show that the error of the track is 1.026‰,which is superior to the classical VO and PDR algorithm,also show the robustness optimization in extreme environment.

关 键 词：行人航迹推算 视觉里程计 反向传播神经网络 扩展卡尔曼滤波 

分 类 号：G434[文化科学—教育学]