Abstract: For case of low recognition accuracy when using universal model to distiguish confusing human behaviors such as walking, going upstairs and downstairs, a mobile user behavior recognition method based on wavelet decomposition was proposed. It extracts the wavelet energy distribution, the number of wavelet peak and the everage wavelet peak amplitude from the sub-signals generated by wavelet decomposition, and also the decision tree classifier is used to build the user-independent behavior recognition model. The typical time-domain feature dataset and wavelet feature dataset were respectively used to train and test the universal model. Experiments show that the proposed method improves the average accuracy about 14.82% of the three confusing behaviors, and reduces the possibility of misjudgment.

Key words: wavelet decompositon, wavelet energy, wavelet peak value, behavior recognition