Team Members: Dr.P.Subashini, CMLI Coordinator,Professor of Computer Science.
Dr.M.Krishnaveni, CMLI Co-coordinator, Assistant Pofessor(SG), Departemnt of Computer Science.
Dr.S.Meenakshi, Associate Professor of Computer Science, Gobi Arts and Science College.
Mrs.S.Bhuvaneswari, Research Scholar,Department of Computer Science.
Ms.Rumaiza Fathima RJ, II MCA , Gobi Arts and Science College.

Project Summary:

Crime detection and prevention have always been critical concerns for society. The increasing number of recorded crimes, including robbery, fighting, shooting, and accidents, highlights the need for advanced surveillance techniques to monitor and analyze crime-related activities effectively. Surveillance videos capture a variety of realistic anomalies, but compared to normal activities, abnormal events are rare. Identifying these abnormal events is a crucial role in video surveillance, requiring AI-based models to enhance crime analysis and response. The Project entitled “AUTOMATED FORENSIC CRIME ANALYSIS SYSTEM USING VISION LANGUAGE MODELS FOR CRIME DESCRIPTION AND SUMMARIZATION IN SURVEILLANCE VIDEOS" aims to develop a robust AI-powered forensic system capable of detecting, describing, and summarizing crime events in surveillance videos. The system integrates VLMs and LLMs to automate crime analysis, reducing the need for manual monitoring. The four important steps which are carried out in this project are: preprocessing, anomaly detection, crime scene description and crime scene summarization. Preprocessing: It involves extracting frames from videos, resizing them to a standard resolution, and normalizing pixel values to ensure uniformity and enhance model efficiency. Frame extraction converts video data into individual images, allowing for frame-wise analysis of crime events. Resizing ensures that all frames maintain consistency across different surveillance sources, making the model more robust to variations in video quality. Normalization scales pixel values within a predefined range, improving model convergence and reducing computational overhead. Anomaly Detection: Anomaly detection is the process of identifying whether a given video contains normal or anomalous events. This process is essential in surveillance systems for automatically detecting suspicious activities without manual monitoring. Crime Scene Description: The crime scene description method generates a detailed textual description of the crime event using a VLM. This description captures crucial details such as the nature of the crime, the actions of individuals involved, and the overall context of the scene. By processing key frames. The generated text follows a structured approach, 6 making it easier to interpret crime events without manually reviewing extensive surveillance videos. Crime Scene Summarization: Crime Scene summarization method utilizes a Large Language Model (LLM) to summarize the extracted description into a concise crime report. This step eliminates redundant details while retaining the most critical aspects of the crime. The LLM processes multiple descriptions from different frames, condensing them into a short, structured summary that provides an overview of the incident. This approach ensures that law enforcement agencies receive clear and actionable reports without having to analyze lengthy textual data.