Violence Prediction System Using Bi-LSTM
P. Sai Kiran, B. Tirapathi Reddy, Goudu Vimal Keerthi, Venkata Subbarao Avanigadda, Chandrika Priyanka Vinnakota, Deekollu Aruna Sri · 2023
Usage of video surveillance to monitor public places has been rapidly increasing. Traditionally, people use human labor to monitor the provinces. To decrease human intervention, one can use computer vision and machine learning techniques for safety monitoring. Generally, the intensity of an action involving violence can be measured from the accuracy of the system used. This work implements an architecture for classifying a video into violent or non-violent, and identifying the violent scenes using MobileNet V2 and bi-directional LSTM. A dataset of 2000 videos are used to train the model, combining hockey fights, movie fights, and real-time fights. The model achieved an accuracy of up to 95%.