Abstract： An algorithm of activity pattern recognition based on the time domain feature of acceleration was proposed for realtime human activity pattern recognition with high accuracy. Time-domain analysis was used as the only method for extracting features. Computation of feature extraction was simplified to achieve realtime recognition of activity, and the ideal result with high accuracy was acquired at the same time. The data tests on Android smart phone indicate that the average accuracy of realtime activity recognition is above 80%. The result proves that this algorithm has a much more high accuracy of realtime recognition compared with the existing algorithms, which has great application prospect in hand-held terminal areas.