3D Convolutional Neural Network for Human Behavior Analysis in Intelligent Sensor Network.

The intelligent recognition of human behavior and action in massive video data is the key application direction in the field of artificial intelligence. With the development of intelligent communication network, multimedia communication has become a hot spot in the field of video analysis. 3D convolution is an efficient deep learning model. It can learn the temporal and spatial features of target images at the same time. A 3D max residual feature map convolution network (3D-MRCNN) is proposed in this paper. Problems can be solved by the proposed model that the deficiencies of the network degradation and gradient disappearance caused by convolution calculation. The proposed model is preprocessed by 2D convolution firstly. A learning network including 3D-max feature map (3D-MFM) and residual structure is established after the convolution splitting is completed. Finally, the output vectors corresponding to the two different inputs are connected and fused into the support vector machine (SVM) classification. The accuracy of 3D-MRCNN can achieve 85.7% by experimenting on the representative UCF101 data set. And it has higher accuracy and operating efficiency compared with the models which have strong correlation with 3D-MRCNN. [ABSTRACT FROM AUTHOR]