Anomaly Detection through Surveillance Videos using Convolutional Neural Network

The footage from the ever-expanding network of surveillance cameras is like an ocean, it contains a large amount of information that is difficult for humans to analyze in real-time. The capacity to analyze and monitor the surveillance cameras is quite limited. In everyday life, suspicious or criminal activities like vandalism, abuse, robbery, stealing, fighting, explosion, etc. occur, detecting such abnormal incidents from surveillance camera videos installed at the corresponding localities and alerting the concerned authorities can help the prevention of mishaps and enhance security. Through this paper, we propose a robust system for anomaly detection in surveillance videos using supervised Machine learning algorithms. The model is built based on CNN (Convolutional Neural Network), using new emerging algorithms like I3D (Inflated 3D) and DenseNet121 algorithms for feature extraction, which makes it possible to gather the most relevant and crucial factors for model training. Through transfer learning, the model has been tested rigorously and improvements have been made through analysis of test and validation scores. The model performance promises increased accuracy, thus fulfilling the objective of creating an accurate anomaly detection system.