Unusual Human Behavior Detection System in Real-Time Video Systems

Abnormal behavior detection, in terms of importance, has become a necessity real-time visual systems. The main problem is the ambiguity difference between characteristics abnormal and normal behavior, which its definition usually different according to previous context images. In this research, three approaches are used. first approach, standard Convolutional Automatic Encoder (CAE) After evaluation, it was found that CAE does not take into account temporal aspect image frames sequence. second method involves automatic encoding learn dataset's spatio-temporal structures. third complex LSTM cells used for further improvement. outcomes test display proposed methods have better performance compared many conventional methods, their efficiency identifying very competitive methods.