Designing an Efficient Framework for Violence Detection in Sensitive Areas using Computer Vision and Machine Learning Techniques
Kuldeep Singh, Kommunuri Preethi, K Vineeth Sai, Chirag Modi · 2018
Human security against violence is one of the major concern in sensitive areas like ATMs, Government offices, Hospitals etc. For such incidents, there is a need of generating timely and automated alerts to concern officials to take further action. For this, surveillance cameras deployed in sensitive areas can be very much helpful. However, existing systems face the problems of low accuracy, high false alerts and high computational cost in monitoring and analyzing the video streams from surveillance cameras and making the decision about violence in real-time. In this paper, we propose an efficient framework to detect violence in sensitive areas using feasible computer vision and machine learning techniques. It collects the video streams of human activities and generates the violence related features by applying motion tracking which slices the video frames based on the presence of moving objects. To find the violent flow descriptors, it calculates the optical flow for each pixel of the frame. These violent flow descriptors are applied to different machine learning techniques to detect the violence events. For the feasibility analysis, different machine learning techniques are analyzed and compared. In addition, the fusion of feasible techniques is tested for improving the accuracy and reducing the error rate.