Pedestrian Navigation Error Correction Method Based On Neural Network And Visual Aid

Pedestrian navigation system is closely related to national security,economic development and people’s life,and plays a key fundamental supporting role in military applications,transportation,firefighting and disaster relief,surveying and mapping exploration and other fields.Pedestrian navigation method based on micro-inertial measurement unit(MIMU)has strong autonomy,high concealment,and can provide continuous positional information,which is of great research value.Pedestrian navigation error correction method based on neural network and visual information assistance are researched in this paper,and the main contents and conclusions are as follows:(1)Inertial navigation algorithm and error equation based on low-cost micro-electromechanical system(MEMS)are studied.The complete and complex strapdown inertial navigation algorithm is simplified by combining the MEMS accuracy index and pedestrian navigation application scenarios,and the error propagation equation of MEMS inertial navigation system is deduced,which provides theoretical support for pedestrian navigation algorithm.(2)To address the problem that the fixed threshold detector cannot perform reliably under different motion patterns,an improved zero-velocity detection model based on neural network is constructed.Firstly,the model structure is built by analyzing the pedestrian foot motion features.Secondly,the model is trained under different motion types using the output labels of the best conventional detector.Finally,the validity of the learned model is verified experimentally.(3)A heading divergence suppression method incorporating visual information is proposed for the problem of unobservable and gradual accumulation of heading errors in zero-velocity update(ZUPT)algorithm of foot-mounted pedestrian inertial navigation system.Firstly,how to extract the camera heading angle from the output information of ORB-SLAM based on binocular vision is studied,then a Kalman filter(KF)model based on heading angle correction in the zero-velocity state of pedestrian is established,and finally the visual heading observation is experimentally verified to effectively suppress the heading divergence.(4)The performance of the improved detection model and the foot heading error correction method incorporating camera pose information are experimentally verified in multiple scenarios.The experimental results show that the improved model can accurately identify the zero-velocity interval under different motion patterns,and the average positioning error is reduced by 32.1% compared with the fixed-threshold conventional detector,and it also has good adaptability in some prior undefined motion types.In addition,the model works well over longer distances and maintains high zero-velocity detection performance in more complex indoor motion scenarios.The foot heading correction method incorporating camera pose information effectively suppresses the divergence of heading angle in straight-line round-trip experiments and rectangular-path walking multi-turn experiments,further improving the solution accuracy of foot-mounted pedestrian inertial navigation system.The research work carried out in this paper enhances the robustness of pedestrian navigation system based on MIMU in response to different motion patterns and improves the heading accuracy of foot-mounted pedestrian inertial navigation system,which has certain reference value and guiding significance for relevant research in the future.