Research On Prediction And Detection Algorithm Of Turning Based On Human Body Motion Characteristics

The human gait perception algorithm has been widely used in gait monitoring,gait rehabilitation,and walking assistance.However,the existing perception algorithms mostly have limitations,that is,only pay attention to the linear gait and the motion in the sagittal plane of the human body,and it is difficult to make the perception of non-linear and asymmetrical gait such as turning gait,which greatly limits the application of gait perception algorithms.The perception of the turning gait mainly includes: the perception of the straight-to-turn transition period,that is,the prediction of the turning gait,and the perception of the turning gait feature,that is,the detection of the turning gait.Therefore,it is of great significance to study the motion characteristics of the turning gait and the prediction and detection algorithm of the turning gait for perfecting the gait perception algorithm and extending the application of the wearable system.Aiming at the problem that the existing gait perception algorithm can not predict and detect the turning gait,this paper studies the human motion characteristics of the turning gait.Based on this,the wearable IMU system layout scheme is proposed,and based on PSO algorithm and the heuristic feature,the turning gait prediction algorithm is proposed.Combined with supervised learning,the turning gait detection algorithm is proposed.The algorithms is optimized and verified by online gait experiment.The following studies have been carried out in detail:Firstly,aiming at the problem of insufficient understanding of the human body's turning gait movement characteristics,the human lower limb kinematics model was established,the gait test was carried out and the experimental data was processed.The kinematic parameters of the main part of the human body and the main joints of the lower limb were obtained.The gait sample library is established.The variance characteristics and the correlation analysis are used to screen out the motion characteristics with significant characteristics during the turn.Then,by comparing and analyzing the curves of these features during the straight walking and turning cycles,the kinetic characteristics of the turning gait in different turning mode were summarized.Secondly,based on the motion characteristics of turning gait,the input of the algorithm and the arrangement of the IMU are determined.The heuristic characteristics of the gait prediction are determined through the PSO algorithm,based on which the three-layer gait prediction algorithm was proposed.The supervised learning(discriminant analysis,decision tree,support vector machine and neural network)was used to train the turning gait detection classifier,through cross-validation method the classifiers was evaluated and the best turning detection model was determined.Finally,in order to scientifically evaluate the performance of the algorithm,an online verification experiment was designed,and the evaluation indexes of accuracy and timeliness were determined.The single-point turning walking experiment and the continuous indefinite turning walking experiment were carried out respectively.The prediction and recognition algorithms were evaluated separately,based on which the IMU layout and the algorithm were optimized separately.Finally,the experimental results showed that the optimized algorithm has better accuracy and speed.