Keyframe Extraction Approaches for Intelligent Video Analysis

Real-time violence detection is essential for protecting the safety and security of people, especially in college campuses that are dynamic and have crowds. Manual surveillance systems are popular but inefficient as they require human attention all the time and are not capable of quickly identifying and responding to violent incidents.Also the exsisting violence detection model takes quite a time for processing the entire video. The proposed model extracts features as keyframes from the video using MobileNet V3 and K-medoids clustering. The redundancy of video data can be avoided and significant violent activities identified by isolating and analyzing critical frames from video data. This enhances the efficiency and responsiveness of surveillance systems. This work primarily contributes a customized violent activities dataset and implementing keyframe extraction algorithm for optimizing the usage of computational resources, and incorporating lightweight algorithms for seamless deployment with existing CCTV networks. The experimental results clearly demonstrate the ability of this approach to detect keyframes, thereby offering a scalable and practical solution.