Gait recognition and analysis is one of the most important biometric methods for medical treatments, virtual reality games and human motion identification. Gait recognition based on wearable MEMS inertial sensors is proposed for medical rehabilitation with Physical Activities Healthcare System (PATHS) in this paper. We use relative wavelet energy as features for support vector machine (SVM) recognition algorithm to discriminate walking pattern from other motion patterns. This method has been proven capable of distinguishing walking gait from other regular physical activities through our experimental validation.