|
AUTHOR(S): Ruba Soundar K, Melwin Prabhu R.
|
TITLE Artificial Intelligence based Automatic Violence Detector in CCTV Footage |
 PDF |
ABSTRACT In this modern society, CCTVs are everywhere to monitor employees, students, and public citizens in malls, hotels, theaters, etc.; Analyzing CCTV videos is mandatory for public security. Many places have cameras, not just one or two, they could have a hundred. Hence it needs human intervention to use it efficiently by checking all the time, which is a time-consuming process and not a much more efficient way to watch them all. Without humans watching, the camera won't be useful, it could just store the videos. This paper proposes a methodology to detect violence by automating surveillance cameras without the involvement of humans by applying deep learning techniques. This work builds a network to detect violence in videos by observing human movements. Proposed network is composed of Resnet50 (Residual Network) in combination with ConvLSTM (Convolutional Long Short-Term Memory). Input frames are processed to extract humans and eliminate the background with the help of computer vision techniques. Upon background elimination, human poses are estimated with the help of MediaPipe solutions, and the frames are passed to the proposed network. Various experiments were conducted on the existing surveillance datasets empirically validate the efficiency of the proposed network by a 5% increase in accuracy compared to State-Of-the-Art (SOTA) violence detection methods. Once the violence is detected, the system will give out a violence alert message automatically thus enabling the respected authorities to take quick actions. |
KEYWORDS Artificial Intelligence, Violence Detection, LSTM, Convolutional Networks |
|
Cite this paper Ruba Soundar K, Melwin Prabhu R.. (2025) Artificial Intelligence based Automatic Violence Detector in CCTV Footage. International Journal of Transportation Systems, 10, 8-13 |
|
