Abstract: In order to improve the accuracy of human activity recognition based on intelligent terminal,we proposed a recognition method based on convolution neural network.We pre-process the raw acceleration data,and input the processed data directly into the convolution neural network to do local feature analysis.After processing,we got the characteristic output items,which can be directly inputted into the Softmax classifier,which can recognize five activity,such as walking,running,going downstairs,going upstairs and standing.By comparing the experimental results,the recognition rate of different experimenters is 84.8%,which proved that the method is effective.

Key words: Human activity recognition,Deep learning,Convolutional neural networks