Research On The Key Technology Of Autonomous Pedestrian Navigation

The autonomous pedestrian navigation system based on micro inertial devices,visual sensors and other equipment can meet the needs of users for long-time and high-precision navigation in indoor,tunnel and other environments where satellite signals are obscured or disturbed.The sensors of the autonomous pedestrian navigation system can independently measure human motion or perceive the external environment without arranging other auxiliary equipment in the target environment in advance.Therefore,the autonomous pedestrian navigation system has good environmental adaptability and anti-interference,and can provide reliable navigation services in extreme application scenarios such as disaster rescue and military action.The autonomous pedestrian navigation technology has important research value and application prospect.The foot-mounted inertial navigation system based on micro inertial devices is a typical autonomous pedestrian navigation system.It adopts the zero-velocity update(ZUPT)algorithm,using the zero-velocity state constraint when the foot contacts the ground.It corrects the navigation error through the extended Kalman filter.The accuracy of zero-velocity detection directly determines the error correction effect of the ZUPT.The traditional zero-velocity detection algorithm uses a fixed threshold to judge the zero-velocity state,which has high detection accuracy in low-speed and regular motion.However,the fixed threshold detection algorithm needs to calibrate the threshold in advance,and the detection accuracy is low in complex and high dynamic motion.In addition,the observability of heading angle error in the system is poor.ZUPT cannot effectively correct the heading error.The pedestrian navigation system based on visual sensor can independently perceive the external environment.Using the geometric relationship of feature points in the image,the pose state of motion can be estimated and the external environment map can be constructed.However,the visual navigation system cannot work well in low texture and dynamic environment.In order to realize long-time,high-precision and high reliability autonomous pedestrian navigation,the main problems faced by the current autonomous pedestrian navigation technology are studied.The main contents of the paper are as follows:1.An adaptive zero-velocity interval search algorithm is designed.According to the characteristics of foot motion,the foot motion is segmented.The "motion-still-motion" is taken as a gait search interval.The inertial data in a gait search interval is mapped into the search space through a nonlinear function.The difference between zero-velocity state data and non-zero-velocity state data will be amplified in the search space.Based on the statistical characteristics of data in the zero-velocity interval and the distribution characteristics in the interval,the zero-velocity interval is adaptively detected in the search space by hierarchical iterative search.2.A zero-velocity detection algorithm based on contrastive learning is designed.Zero-velocity detection is essentially a binary classification problem,so it can be designed based on deep learning.Static is essentially a relative state,which is obtained by comparison with other reference objects.Contrastive learning is similar to this method.It can determine the zero-velocity state by comparing with the reference data.Firstly,the algorithm eliminates some inertial data that are absolutely impossible to be in the zero-velocity state.It only trains and determines those data that are more likely to be in the zero-velocity state,and then compares the inertial data with the reference static data to determine the static state.The amount of training data required for contrastive learning is smaller,and the applicability of neural network on non-training data sets is better.3.The error correction algorithm of inertial pedestrian navigation is studied.The correction process of zero-velocity correction algorithm is discussed.The factors affecting the covariance matrix in the filtering process are deduced.The effects of measurement performance,sampling frequency and installation position of micro inertial system on the correction effect are analyzed.This can provide guidance for the design of foot bound inertial pedestrian navigation system.Since ZUPT cannot correct the heading angle error,the heading correction is carried out based on foot motion constraints and external magnetic information constraints4.The navigation system of vision/foot inertial multi-source information fusion based on factor graph optimization is studied.In order to improve the accuracy and reliability of autonomous pedestrian navigation system,a multi-sensor fusion algorithm framework of autonomous pedestrian navigation based on factor graph optimization is designed.The framework can expand sensors according to the actual system needs,and take the advantages of different sensors.Based on this framework,the vision/foot inertial multi-source information fusion pedestrian navigation hardware system which can carry out real-time navigation is built to realize accurate and reliable autonomous pedestrian navigation.