Anomaly detection in video surveillance over the crowded environments for security support

Abnormality detection including human unusual movement, unusual traffic, crime sense, group violence, etc. is inspirited by virtual data availability which is obtained from closed circuit television (CCTV). The logic of abnormality existence is made in assumption that most of the virtual data (surveillance data) is not included abnormal events with 100 % probability. The abnormal event is less probable than the normal one hence, appointing of people for surveillance who observe the data for log times is no longer appreciated due to the time and man power wastage. Automatic abnormality detection is proposed in this work using deep learning technology to perform the detection and dispense man power surveillance which ensures good performance is far less expensive budgets. Long short term neural network (LSTM) is used for the same; LSTM performance is compared with the proposed state of the art e.g. Feed Forward Neural Network accompanied with K-nearest neighbour particle swarm optimization (KNN-PSO). The proposed state of the art is outperformed in abnormality detection accuracy, the maximum recognition accuracy was 99.18345128 percent.