| 摘要: |
| 针对传统的行人航迹推算（Pedestrian Dead Reckoning,PDR）方法无法满足多运动状态下的定位问题，提出了一种基于神经网络运动识别辅助室内定位的方法。构建出卷积神经网络(Convolutional Neural Network,CNN)和门控循环单元（Gated Recurrent Unit,GRU）组合的神经网络模型，用于识别人体的运动状态并完成分类。根据运动分类的结果应用到行人航迹推算中，分析和筛选运动参数特征作为算法的阈值约束条件来提高定位精度。在算法中运动步数由合加速度计数据波形检测得到，步长由运动状态的特征自适应调整步长模型。通过实验验证，CNN-GRU模型在自建数据集上的准确率达到99.6%。将识别结果应用到PDR中，在112 m 4种动作的矩形路线中定位误差为1.8 m，误差远低于传统PDR定位的19.9 m。实验结果验证了该方法的可行性。 |
| 关键词: 室内定位 行人航迹推算(PDR) 人体运动识别 卷积神经网络(CNN) |
| DOI：10.20079/j.issn.1001-893x.221212001 |
|
| 基金项目:国家重点研发计划（2020YFC1511702）；北京市自然科学基金（4212003）；北京市科技新星计划交叉学科合作课题（202111） |
|
| An Indoor Localization Method Based on Human Motion Recognition Constraints |
| LI Jiazhi,LIU Ning,JIE Xiaohan,WANG Jingxiao,ZHAO Hui |
| (Beijing Key Laboratory of High Dynamic Navigation Technology,Beijing Information Science and Technology University,Beijing 100192，China) |
| Abstract: |
| For the problem that the traditional pedestrian dead reckoning(PDR) method can not satisfy the location in multi-motion state,a new indoor localization method based on neural network motion recognition is proposed.A neural network model combining convolutional neural network(CNN) and gated recurrent unit(GRU) is constructed to identify the motion state of the human body and complete the classification.Then,according to the results of the motion classification,it is applied to PDR,and the motion parameter features are analyzed and screened as the threshold constraint conditions of the algorithm to improve the positioning accuracy.In the algorithm,the number of moving steps is detected by the waveform of the combined accelerometer data,and the step size adaptively adjusts the step size model by the feature of the moving state.The experimental results show that the accuracy of CNN-GRU model on the self-built data set is 99.6%.Then the identification results are applied to the PDR.The positioning error of the rectangular route with four movements of 112 meters is 1.8 meters,which is much lower than the 19.9 meters of the traditional PDR positioning.The experimental results verify the feasibility of the proposed method. |
| Key words: indoor localization pedestrian dead reckoning(PDR) human behavior recognition convolutional neural network(CNN) |
