Deep Neural Network based Anomaly Detection for Real Time Video Surveillance

One of the main concerns across all kinds of domains has always been security. With the crime rates increasing every year the need to control has become crucial. Among the various methods present to monitor crime or any anomalous behavior is through video surveillance. Nowadays security cameras capture incidents in almost all public and private place if desired. Even though we have abundance of data in the form of videos they need to be analyzed manually. This results in long hours of manual labour and even small human discrepancies may have huge consequences negatively. For this purpose, a Convolution Neural Network (CNN) based model is built to detect any form of abnormal activities or anomalies in the video footages. This model converts the input video into frames and detects the anomalous frames. To increase the efficiency of the model, the data is de-noised with Gaussian blur feature. The avenue dataset is used in this work to detect and predict various kinds of anomalies. The performance of the model is measured using classification accuracy and the results are reported