Anomaly Detection in Video Events using Deep Learning

The anomaly detection system gives a solution to detect anomaly in crowd event video and sets alarm for public safety in mass gatherings. The deep learning technique CNN(Convolution Neural Network) is used to detect anomaly at the initial stage from the input video and set alarm to avoid damages. The proposed system gets frames from input crowd video to detect anomaly activities are namely fighting, running, protesting, and firing. If any one of the anomaly namely fire, fight, protest and run is occurred in a video, that anomaly is detected from specified frames of video. The specified frames are extracted from a video to find the location of the anomaly. The anomaly detection system makes an alarm sound for the specified location of the anomaly. Using a GSM module, the system sends messages to the controller of the fired area. In the existing system, they used sensors and board for finding the fire. Thus, the proposed system detects the anomaly on video using computer vision based deep learning technique. Thus the anomaly detection system provides simple web camera with alarm for public safety with less cost compare with others