Self-Adaptive Recognition Of Human Gait Based On Improved Hidden Markov Model

Human gait recognition is the technology that using the theories of kinematics,signal processing and pattern recognition to analysis and process the kinematics and physiology gait signals obtained by sensors.It has been widely used in humanoid robot,man-machine coupling robot(exoskeleton and artificial limb),medical diagnosis,rehabilitation,kinematic analysis,identification and authentication.Human gait is quasi-periodic.The transition between the gait phases can be seen as a Markov chain.As there is a Markov chain,so the theory of HMM(Hidden Markov Model)has been widely used in gait recognition.But there are two defects while using HMM dealing with gait recognition.One is that HMM using statistic methods to describe residence time distribution function,it can't describe the time response of human gait very well.The other is once the training is done,the parameters of HMM is fixed,so it can't adapt with different situations.These defects limits the recognition rate of human gait.In this article,we use the acceleration signal of thigh to build a five segments gait model.With the improvement of Hidden Markov Model,the recognition performance of gait phases and the adaptive capacity of gait model have been promoted.Here are the specific work.1.The gait data has been pre-processed and the feature has been extracted.The pre-processing parts includes abnormal data deletion,denoising and normalization etc.The feature extracting parts includes splitting data into windows and extracting feature form each single window.2.The HMM theory has been discussed.The defects of HMM used in human gait recognition has been analyzed and we indicate the way to update HMM.3.We bring time parameters into HMM and replacing self-transfer probability with residence time distribution function to describe human gait better.4.We do some improvements on HMM to make it has the ability of adapting different human,different action and different environments.Self-adaptive algorithms has been brought into HMM to make it more robust.5.We do some improvements on the issue of single reference model in self-adaptive.We combine gait recognition and action recognition together to reach the goal of automated choosing reference model.Finally,we do some experiment to verify the effects of those improvements.The results show that these improvements can rise the recognition rate of human gait and the fixed model has the ability of adapting different human,different action and different environments.