基于行为概率分析的室内行人导航方法

Indoor pedestrian navigation method based on behavior probability analysis

行人在室内场景下进行混合步态运动时,传统零速检测方法易出现“漏检”和“误检”,造成定位精度下降。针对此问题,提出了一种基于行为概率分析的行人零速检测与爬楼高度估计方法。一方面,设计了基于步态周期检验的最大概率零速点提取方法,通过足部零速广义似然比概率统计对步态周期进行准确划分,继而实现对步态周期内最大概率零速点的提取;另一方面,采用基于长短期记忆的深度学习网络模型实现上下楼运动检测,基于气压高度计与惯性器件数据,构建描述足部跨越台阶数的能量概率函数,并对台阶高度、跨越台阶数以及爬楼高度进行实时在线估计。最后开展了行人室内导航实验,相较于传统零速检测算法和零速阈值自适应调整算法,所提算法的平均零速点提取成功率分别提高24.44%和16.21%,平均导航定位性能分别提高了55.14%和39.15%,高度误差由7.79%降低到0.39%。

When pedestrians perform mixed gait movement in indoor scenes,the traditional zero speed detection method is prone to"missing detection"and"false detection",resulting in the decline of positioning accuracy.To solve this problem,a method of pedestrian zero-speed detection and height estimation based on behavior probability analysis is proposed.On the one hand,a method of extracting maximum probability zero speed points based on gait cycle test is designed.The gait cycle is accurately divided by the generalized likelihood ratio probability statistics of foot zero speed,and then the maximum probability zero speed points within the gait cycle are extracted.On the other hand,a deep learning network model based on long and short term memory is used to detect the movement up and down stairs.Based on the barometric altimeter and inertial device data,the energy probability function describing the number of steps crossed by the foot is constructed,and the real-time online estimation of the step height,the number of steps crossed,and the climbing height is carried out.Finally,indoor pedestrian navigation experiments are carried out.Compared with the traditional zero speed detection algorithm and zero speed threshold adaptive adjustment algorithm,the average zero velocity point extraction success rate of the proposed algorithm is improved by 22.44%and 16.21%,the average navigation positioning performance is increased by 55.14%and 39.15%respectively,and the height error is reduced from 7.79%to 0.39%.