A Hybrid Novel Methodology for Human-Detected Keyframe Extraction in Crime Scene Analysis

Abstract

Introduction: Surveillance system research is currently growing rapidly. Residential, public transit, financial institutions, and other public significant locations where it is frequently employed. It is also crucial for safeguarding critical infrastructure. Due to its numerous uses, human identification in surveillance system video scenes has become more and more popular in recent years. For a number of application areas, including crime scene investigation, a video surveillance system's ability to accurately recognise humans is crucial.

Objectives: This paper presents the Human Detected Keyframe Extractor (HDKFE), a novel framework for efficient keyframe extraction tailored to crime scene investigation and surveillance applications.

Methods: HDKFE integrates Faster R-CNN with an optimized threshold tuning and Keyframe extraction using Local Maxima with Canny Edge Detection selectively capture frames containing high human activity, effectively reducing redundancy while preserving critical content.

Results: Comparative evaluations on existing and real-time surveillance datasets demonstrate HDKFE’s superior performance, achieving compression ratios up to 99.22% with high accuracy and also a comparative with different techniques HDKFE method proves with 98.94% higher accuracy.

Conclusions: This efficiency reduces storage demands and enables streamlined analysis of extensive footage, positioning HDKFE as an effective tool for evidence gathering in dynamic surveillance environments. Future advancements may focus on real-time adaptability, integration with predictive analytics, and improved automation to support comprehensive monitoring and proactive security measures.