The Research Of Pedestrian Behavior Tracking Based On Inertial Sensor

In recent years,pedestrian behavior tracking technology has become a hot research issue in the field of pattern recognition and artificial intelligence.Compared with the visual,electromagnetic,mechanical,and acoustic tracking systems,inertial systems have been widely concerned due to the advantage of low cost,small size,also avoiding complicated auxiliary equipment and harsh environmental requirements.As the complexity of human motion and the fact that the inertial sensing unit is susceptible to noise interference and some drift error,it is difficult to implement behavior tracking in complex motion modes without external reference points.This thesis designed a root node displacement algorithm in multiple motion mode to track pedestrian behavior.First of all,this thesis is based on the introduction of the basic principle of pedestrian behavior tracking,system composition and the functions of each module,which establishes a rigid body hinge model of human motion.Aiming at issues as large noise,poor dynamic performance,and low accuracy of sensor parts,this thesis proposes a solution based on Quasi Newton's Adaptive Fuzzy Logic Complementary Filtering Algorithm to solve joint posture information during motion.In order to verify the reliability of the algorithm,static experiments and dynamic experiments are designed in this thesis.The results show that the root mean square error of the system is less than 0.04°in the stationary condition,and the root mean square error of the system is less than 1.3°in the dynamic condition.Secondly,three kinds of behavior patterns,walking,running,and jumping,which are common in daily life,are selected as research objects,and the corresponding acceleration time domain eigenvalue is extracted to realize behavior recognition.According to the change of foot acceleration during running,it proposed how to distinguish between the leap period and the support period,and algorithm of the support leg in the support period.Combined with the characteristics of the three movement modes and the joint attitude information,the displacement estimation of the root node in the multiple motion mode was realized.And use of the volume Kalman filter to compensate the displacement estimation results to improve the system accuracy.The experimental results show that the root mean square errors of walking,jumping,running is less than 1.261 dm,1.054 dm and 1.826 dm in the three motion modes;the accuracy after compensation is increased by 50.3% and 57.33%,47.9% respectively if compared with that before compensation.The offset displacements of the flat and uneven pavement in the hybrid model are within 3.9% and 4.2% respectively,which can accurately depict the human body position information and posture information in the global coordinate system.Finally,on the self-designed inertial pedestrian behavior tracking platform,we use motion data to drive the virtual character and reproduce the body's movement posture to verify the feasibility and reliability of the pedestrian behavior tracking system.