Deep learning techniques for physical abuse detection

<span>Physical abuse has become a societal problem. Mostly children, women and old age people are vulnerable to it especially in cases of domestic violence or workplace aggression. Reporting is itself challenge if there pre-existing relationship between the abuser victim. In this paper we propose deep learning technique for human action recognition pose identification tackle physical by detecting real time. 3D convolution neural network (CNN) architecture built using feature extractors which extract both temporal spatial data video. With multiple layer subsampling layer, input video been converted into vector. Human estimation done detection key points on body. Using these tracking them from one frame another gives spatial-temporal features feed (NN). We present metrics measure accuracies such systems where time reporting fault tolerance capabilities utmost importance. Weighted shows accuracy about 89.42% with precision 85.82% thus effectiveness system.</span>