Research On Gait Recognition Based On Multiple Sensors Information Fusion

In recent years,active smart bionic leg has attracted great concern in the field of rehabilitation medical robots.With the development of biomedicine,the application of surface myoelectric signal(s EMG)has become more and more mature.As one of the core technologies for bionic legs,gait recognition technology is of great significance for improving intelligence of the bionic legs.Existing gait recognition lacks the interaction of bioelectrical signals and cannot respond to human movement intentions more quickly.There are problems such as low robustness and high feature redundancy in fusion of multiple information for gait recognition.In this regard,this article carried out the following research work:(1)Design of gait information collection system and data collection experiment.s EMG signal was collected by Myomove electromyograph and transmitted to the host computer through wireless routing mode.The JY901 attitude instrument is used to collect acceleration,angular velocity and angle signals,and the Flexi Force thin film rheostat and ADS1256 IDB analog-to-digital conversion module are used to collect pressure signals.Both physical signals are transmitted to the host computer through a serial port.Finally,codding with Lab View to realize sensor communication,disintegrate and display and storage data.In the signal collecting experiment,10 experimenters’ movement data was collected for five gait patterns of walking on the ground,uphill,downhill,upstairs and downstairs.(2)Signal noise reduction processing,feature extraction and selection.Second-order Butterworth filtering is used to obtain the 10-450 Hz frequency band signal.A variety of mode decompositions combined with wavelet threshold are compared,and variational mode decomposition combined with wavelet threshold performs better effect for noise reduction of the signal.The time domain,frequency domain,time-frequency domain,multiple types of entropy characteristics and the maximum Lyapunov index methods were used to extract the signal features,and the Davies-Bouldin index was used to analyze the spatial separability and spatial distribution of signal was used to analyze which features is more conducive to distinguish gait,and the maximum correlation and minimum redundancy method based on mutual information was used obtain a better combination feature.(3)Research on gait recognition algorithm.The existing random forest(RF)algorithm has the disadvantages of the same weight of the weak classifier and the difficulty of multi-parameter tuning.In this paper,the weight of the weak classifier of the RF algorithm is adjusted by the accuracy of the data outside the bag and the cuckoo optimization algorithm to obtain a better combination of parameters.The traditional cuckoo algorithm uses a fixed discovery probability and Levi ’s flight to update the bird ’s nest,which has the disadvantage of slow convergence.The proposed adaptive cuckoo(ACS)algorithm,which adaptively adjusts the discovery probability and Levi ’s flight update step size,improves the efficiency of searching for reaching convergence.Finally,the characteristics of different types of signals was identified and analyzed for comparing the effects of gait recognition,and the s EMG signal features of medial femoral muscle and rectus femoris muscle and the features of the posture signal fixed on the thigh was selected for multi-gait recognition can obtain better effects,the recognition accuracy rate reaches about 93%,which meets certain accuracy requirements.