Enhancing Public Safety Through Intelligent Video Analytics For Violence Detection

Abstract

Surveillance Cameras Are Becoming More Prevalent Worldwide Thanks To Advancements In Digital Video Technology, But The Sheer Amount Of Footage Generated Makes It Tough For Humans To Analyze In Real-time. Even Manual Efforts Can Lead To Delays In Spotting Events. However, With The Rise Of Machine Learning, Tasks Like Automatically Detecting Violence In Videos Have Become More Feasible. Our Study Delves Into Using Intelligent Networks That Use 3D Convolutions To Understand The Changing Relationships Between People And Objects Over Time, Capturing Both The Spatial And Temporal Aspects Of The Data. We Also Make Use Of A Pre-trained Model For Action Recognition To Enhance The Efficiency And Accuracy Of Violence Detection In Surveillance Videos. To Validate Our Methods, We Assess Them On Various Public Datasets Containing Diverse And Challenging Video Content. Our Experimental Results Show That The Proposed Method Achieves A Substantial Improvement In Detection Performance, Increasing Accuracy From 75% In Existing Systems To 91%. This Demonstrates The Effectiveness Of Our Approach In Handling Complex Real-world Surveillance Scenarios While Maintaining A Lightweight And Efficient Model Architecture Suitable For Real-time Applications.

Keywords