Indoor Pedestrian Multi-motion Pattern Recognition Methods Based On Multisensor Fusion

In recent years,the technology and system of indoor pedestrian autonomous location based on MEMS inertial sensor have been paid more and more attention.Accurate and real-time identification of pedestrian movement in the room has an important auxiliary role in improving the precision of inertial positioning system.Based on this background,the research on indoor pedestrian motion pattern recognition technology based on multi-MEMS inertial sensor information fusion is carried out in this paper.The main research work is as follows:(1)In this paper,the main types and characteristics of indoor pedestrian motion patterns are deeply analyzed,and the key problems to be solved in accurate and real-time identification of pedestrian motion patterns are discussed,combined with the application background of indoor pedestrian inertial positioning.The scheme of multi-motion pattern recognition for indoor pedestrians based on foot-bound MEMS inertial sensor is established:The MEMS sensor with accelerometer and gyroscope is used to collect the inertia data of foot of five normal indoor pedestrian motion modes.Based on the characteristics of foot motion,the data are effectively segmented and extracted.The information of accelerometer and gyroscope is fused to recognize the motion pattern to improve the recognition accuracy.(2)The multi-sensor data fusion model based on feature level fusion is constructed.Combined with the application background,the characteristics of multi-sensor pedestrian motion pattern recognition are determined: seven acceleration time domain features and six gyroscope attitude features.The extraction methods and algorithms of various feature quantities are discussed.(3)In view of the difficulty of window length selection in the traditional method of extracting feature quantities by using fixed-length sliding window intercepting data,the difference of feature quantities in different modes is not significant enough.The low temporal resolution of feature is not conducive to real-time recognition of motion pattern.A method of data interception and feature extraction based on single-step partition is proposed.Due to the periodicity of foot motion,the inertial sensor data obtained by the foot-bound sensor will show the same periodicity.Based on this periodicity,the inertial sensor data ofhuman body motion is segmented in one step,then the features of each single step data are extracted,and the motion pattern is classified and recognized by support vector machine(SVM).The analysis shows that the feature extracted by the single-step partition method has a strong representation of motion,and the difference is significant.It has a strong real-time performance in the multi-motion pattern switching recognition of pedestrians.Finally,the experiment scheme is designed to validate the method of indoor pedestrian motion pattern recognition proposed in this paper.The results show that the single-step feature method can significantly improve the accuracy of pedestrian motion pattern recognition,and the average accuracy is increased by 6.8%,compared with the common sliding window method for data capture and feature extraction.At the same time,the real-time performance of single-step feature method is better than sliding window method.Compared with the method of motion pattern recognition using a single accelerometer,the method of multi-sensor information fusion is more accurate to recognize the daily motion pattern of indoor pedestrians.The recognition accuracy is improved from 93.27% to 97.74%.