Anomaly Detection in Video Surveillance

Abstract

In this project, we have developed an anomaly detection system which makes use of Machine Learning to detect anomalies to include Violence, Theft, Accident, Arson, and Abuse. This would be accomplished by using deep neural networks. Two approaches have been adopted to fulfil the requirement, Multiple Instance Learning approach and MobilenetV2 approach. Both use convolutional neural networks to train machine learning models on different datasets. Both approaches have been implemented on different platforms with different frameworks. Predesigned datasets have been used for training models, datasets comprise of large number of videos containing normal and anomalous behaviors. After training on different frameworks, models are designed to detect anomalies from real world scenarios. Both models have been rigorously tested and display high accuracy for the anomalies mentioned above. Model is then deployed on an interface which takes the video as an input and displays results, either as a graph or in the form of video depiction as per the requirement, for our different approaches. The output extracted can further be utilized for deployment on the end system for real-time anomaly detection in surveillance videos.