Gait Recognition Based On Multi Source Information Fusion

Gait is people’s walking style. Different people have different walking style. Gait recognition is a new field for the biometric recognition technology. Its aim is to recognize people or detect physiological, pathological and mental characters by their walking style. Gait recognition technology is gradually playing its role in intelligent monitoring, clinical, rehabilitation, movement analysis, design of intelligent artificial legs, and many other areas of recognition, it has become the second-generation biometric recognition technology representatives. In order to improve the accuracy of human lower limb gait recognition, a gait recognition approach for the lower limb was proposed in this paper. It was based on multi-source information. This paper investigate the usefulness of different data sources commonly suggested for user gait recognition、signals preprocessing、 feature extraction、 feature fusion、gait recognition. The main research content is as follows:This study investigated the use of surface electromyography(sEMG)、accelerometer、plantar pressure signal and gyroscope combined with pattern recognition(PR) to identify user locomotion modes. The approach infers user’s intents of upslope, downgrade, stairs ascent, stairs descent or level-ground walking. This study investigated the influence of walking speed on the gait parameter、hip joint angle and lower limb muscle group. It can provide data support in order to know users’ body movement.The sEMG signals were instability、randomness、weak. The sEMG signals were polluted in the process of acquisition. sEMG signals were filtered by the weighted moving window filter. The mechanical signals were filtered by wavelet denoising filter. The sensor data streams utilized for the gait recognizer were chosen to reflect the state of the gait. Appropriate sensor information includes electromyography、joint angles and angular velocities of the hip joints, in addition to measured interaction forces. In this paper, features were extracted from frames of data.In order to decrease the time required to train the models, to prevent overfitting, and to facilitate real-time implementation, the feature space was reduced(at the cost of information content). Recent study in myoelectric pattern recognition also indicates that principal component analysis(KPCA) dimension reduction improves classification accuracy for a similar problem. Though multiple possibilities exist for such dimensional reduction, two effective approaches include PCA and KPCA. Both approaches employ linear transformations, which facilitate computational efficiency due to matrix multiplication operations. In this paper, both approaches were considered. But KPCA is better than PCA.Signals are subsequently reduced to a lower dimensionality(for computational efficiency) using KPCA. These data are initially used to train FOS models, which classify the patterns as upslope, downgrade, stairs ascent, stairs descent or level-ground walking. The new features are input variables of the FOS model. The trained models are subsequently used to infer the user’s intent.A pattern recognition method for intent recognition based on the multi-kernel multi-class relevance vetor machine was proposed. The method integrated information of different gait sources with different kernel functions. Particle swarm optimization algorithm was utilized for optimal design to obtain the optimal combination parameters. Experimental data of gait indicated that the designed classification model integrated various feature information and could represent gait characteristics comprehensively with higher accuracy.