Gesture is a compelling user interaction modality for enabling on-the-go interactions.With the usage of the acceleration sensor is becoming more com mon,gesture recognition will play a huge role in human-computer interaction.However,for "continuous" gesture recognition,there is an important problem that how to get the data we indeed need in a series of raw gesture data.To solve this problem,we present an efficient gesture recognition pipe line,which is based on the Hidden Markov Model(HMM).The optimized gesture models are obtained by using K-Means algorithm and moving average filtering.Its optimized for "continuous" recognition while minimizing pro cessing overhead and enhancing usability.The pipeline is constructed to allow for early filtering of unwanted sensor data with minimal processing cost.In this work,the pipeline uses a method of automatically intercepting gesture data to solve the problem that the user is required to delimit explicitly the start and end of gestures.The final result demonstrates that the pipeline can a chieve considerable processing and power saving without impacting overall recognition accuracy.