Research Projects

Rationale of the Research: Surveillance cameras are crucial for public safety, especially in Indian states where timely crime detection and prevention are vital. Current systems face limitations due to the manual, time-consuming review of extensive video footage, which is prone to human error and delays critical security responses. This research proposes an automated solution using Content-Based Image Retrieval (CBIR) to enhance real-time, targeted video retrieval through automated analysis based on user image queries, boosting situational awareness and response efficiency. Scientific Objectives: 1. Develop an automated video retrieval system: Create a mobile or web application leveraging CBIR for retrieving specific video segments from surveillance footage. 2. Enhance object detection and feature extraction: Utilize YOLO v9 for advanced object detection and genetic coding techniques for high-accuracy feature extraction. 3. Integrate relevance feedback: Implement user feedback mechanisms to iteratively refine and improve search results. 4. Improve surveillance efficiency: Demonstrate reduced time and effort for operators in locating relevant footage, enhancing public safety. Hypothesis/Model to be Tested: The hypothesis is that an intelligent surveillance system based on CBIR with genetic deep learning and user feedback can outperform traditional manual video analysis in terms of speed and precision, enhancing the accuracy and efficiency of video retrieval. Main Experiments to be Carried Out: 1. Object Detection in CCTV Video: Convert video footage into frames using the Group of Pictures (GOP) technique and apply YOLO v9, pre-trained on the COCO dataset, to detect and classify objects. Store coordinates and cropped images of detected individuals in a structured database. 2. Feature Extraction and CBIR Application: Apply genetic coding to convert images into amino acid sequences based on Watson-Crick rules. Construct feature vectors from these sequences’ probability occurrences and develop a similarity matching algorithm using Euclidean distance to identify top-k matches. 3. Relevance Feedback for Video Retrieval: Present top-matched images to users for feedback to refine results and retrieve video segments based on coordinates and timestamps, displaying relevant footage with bounding boxes. 4. Mobile/Web Application Development: Build a user-friendly interface enabling image uploads and CCTV video selection, with real-time alerts, video saving/sharing features, and relevance feedback for continuous optimization. Project Significance: This project could significantly advance video surveillance technology, offering a fast, automated solution for law enforcement and security operations. By improving retrieval efficiency, it supports rapid response and strategic focus. Beyond surveillance, potential applications include traffic monitoring and search-and-rescue missions, underscoring its broader impact on public safety and multimedia analysis.