Abstract
There is a need for developing deep learning solutions to analyze videos to
identify any violence being present. This paper proposes a method for the detection of
the presence of violent activities in videos using Deep Neural Networks. Recently there
has been a rapid development happening in the field of Deep Neural networks, but the
number of solutions that have been developed for violence detection is very few. The
proposed solution will play a major role in transforming the way law enforcement
works and support the government’s initiative to make cities smarter. The model is
built using CNN for video frame feature extraction and LSTM to capture localized
features present in the video frames. The LSTM extracts the localized features using
the spatiotemporal relationship between the video frames. The local motion present in
the video is analyzed. This work focuses on accuracy and fast response time. The
performance was evaluated on the hockey fight dataset to detect violent activities.