Two-Stream CNN Architecture for Anomalous Event Detection in Real World Scenarios

Abstract

Development of intelligent surveillance systems for anomalous event detection is essential to save human life and property. In recent years, various deeplearning techniques have been proposed to increase the system performance. However, due to complexity involved in the deep architectures very few can be applicable to real world scenarios. In this paper, a two-stream CNN architecture for anomalous event detection is proposed by exploiting both normal and anomalous videos. Furthermore,a database pre-processing algorithm is proposed to capture the spatial and temporal information for each second in a video, which is given as input to the two-stream CNN architecture. A standard dataset, UCFcrime dataset is used for the validation of the proposed method. Finally, a comparison is made with several existing deep learning approaches in terms of accuracy and it is evident that the proposed method is superior to other state-of-the-art techniques in terms of classification accuracy.