Research On Recognition And Prediction Algorithm For Lower Limb Motion State

With the advancement of science and technology and the improvement of standards inevery professional field, the research work of prosthesis also enters a new stage. It isbecoming not only the simple supporting but also comfortable and intelligent. At present,the active transfemoral prosthesis becomes the focus of research. It can provide assistanceaccording to the gait, can reduce the wearer’s energy consumption in walking process, andcan adapt to more complex walking environment. How can it become effective for gaitrecognition and prediction is the difficulty in the research for active transfemoral prosthesis.Based on these problems, three walking pattern recognition methods and gait predictionalgorithms are presented in this paper. These methods can predict for lower limb movementbased on the date which collected from the sensors. All of these results can providetheoretical foundation for the control of prosthesis.Firstly, acceleration sensor, gyroscope and plantar pressure sensor are used to collectinformation of five kinds of walking pattern (walking, upstairs, downstairs, uphill,downhill). According to gait cycle extracting the data of single cycle, gait phase of thesingle cycle are divided. And then the features of sensor signal under different gait phaseare extracted, which are prepared for gait recognition.Secondly, three kinds of identification algorithms are proposed, whose inputs are thesensors features. These identification algorithms are used to identify the test samples afterthese networks be trained. From the recognition date, the FOS algorithm is proved to havemore higher accuracy.Finally, prediction model is established by three kinds of recognition algorithms, whichis used to predict the joint angle. The input of the prediction model is joint angle andacceleration feature, the output is joint angle. Compare the prediction result and error, theOIF Elman algorithm have the best prediction performance.