Gait Recognition Based On Probabilistic Neural Network

With the development of technology and living standard, powered prosthetics willgradually replace the passive prosthetics because the advantage of powered prosthesis isautomatically changing torque or other parameters according to the external environment.Therefore, powered prosthetic is able to identify walking gait of persons who wearprosthetic and provide the impetus and these are its advantages, but these points are alsodifficult points in prostheses control. The issue proposed the use of Probabilistic NeuralNetwork to identify gait, constituted the input vector by screening gait characteristic valuesof multi-sensor information, completed the identification of gait after walking through theSwing Phase to Initial Stance, and provided accurate information to control the poweredprosthetic.Firstly, use Vicon MX system (capture video signals by camera) and sensor systems(including accelerometer, gyroscope and plantar pressure sensor) to collect five kinds ofgait information. Video signal can be a standard signal to correct sensor information. Byanalyzing the characteristics of human gait, analysis the signals according to the gait cycle,extract feature and screen by using the method of Mean Impact Value (MIV) and PartialLeast Squares (PLS), after screening we can get the feature matrix as the input layer of theneural network.Then, research on the gait pattern recognition algorithm. After comparing ProbabilisticNeural Network (PNN) to Radial Basis Function Neural Network (RBFNN) andBack-Propagation Neural Network (BPNN) in recognizing gait, conclusion is that PNN isbetter than others. So, design two kinds of improved PNN network structure: combine apriori variable and PNN network, dual PNN network.Finally, identify gait using the neural network as above. Contrast recognition time andrecognition rate, we can conclude that using network which is combined a priori variableand PNN network has short recognition time and high recognition rate.