Crime Scene Analysis Using YOLOv5 Model

Crime scene detection and analysis are two critical aspects of real-time surveillance systems. To compete with the increasing crime incidents, timely detection of crime, its suspects and victims is of great importance. This also includes timely forensic investigations and proper record-keeping. This work presents a novel approach to handle these two aspects of crime scene identification and analysis by creating an automated system. The work uses object detection techniques to facilitate real-time identification of crime scenes, automated evidence collection and record-keeping, an alert generation feature, along with a generative text for crime scene description and summarization. To achieve crime scene identification, the most popular You Only Look Once (YOLO) deep object detection model has been optimized. For automated evidence recording, scene analysis is carried out and objects present in the scene are recorded to help further investigations. The paper investigates the performance of our enhanced YOLO model in accurately identifying and localizing objects in various crime scene scenarios, highlighting its potential to enhance the efficiency and accuracy of forensic investigations.