Residual Attention based Long Short-Term Memory for Anomaly Human Behaviour Detection

The recognition of video anomaly is an effective computer vision task which played an essential part in smart surveillance and safety of public however that is challenging because of their complex in real-time surveillance. Different Deep Learning (DL) algorithms utilize essential number of training information without the ability of generalization and with high complexity of time. In this research, proposed a Residual attention based Long Short-Term Memory (Residual based LSTM) method for anomaly human behaviour detection. The UMN dataset is used for detecting human behaviour and it is pre-processed by Adaptive Median Filter (AMF) and Histogram Equalization (HE). Then the pre-processed data are extracted by Residual Network 101 method which extracts essential features and then features are given to Residual based LSTM which detected the anomaly human behaviour. The evaluation parameters used for evaluating a proposed method are accuracy, precision, recall and f1-score. Proposed algorithm reached 98.71% accuracy, 98.63% precision, 98.36% recall and 98.01% f1-score which is effective than other existing algorithms like Robust technique and Spatio-Temporal descriptor.