Background

CCTV cameras have become crucial assets in investigating crimes, accidents, and missing person cases. However, investigators often face the tedious and time-consuming task of manually reviewing hours of footage from multiple cameras to locate specific persons, vehicles, or objects of interest.

There is an urgent need for an AI-powered video analysis tool that can automatically detect and extract relevant frames based on searchable keywords (e.g., 'red car', 'person in black jacket') or reference images (e.g., suspect’s photo). Such a tool would significantly reduce investigation time and improve operational efficiency.

Problem Statement

There is currently no unified, user-friendly tool available to investigators that can process large volumes of CCTV footage and automatically extract relevant frames or segments based on visual similarity or contextual keywords, especially in field investigations with limited technical resources.

Key Objectives

• Assist investigators in quickly locating critical moments from long CCTV videos
• Improve accuracy and reduce manual effort
• Enable rapid response in missing person, theft, or terrorism cases
• Provide forensically sound and exportable output for case documentation

Functional Requirements

Design and build an AI-based CCTV Analysis Tool (desktop or web-based) with the following core capabilities:

Core Functionalities:

1.  Input Handling:

o Ingest raw CCTV footage (common formats like .mp4, .avi, .mov)

o Allow batch upload from multiple cameras or timeframes

2.  Search by Reference Image:

o Upload a photo of a person, vehicle, or object

o Perform face/object re-identification across frames

o Return timestamped frames or short video clips where matches are detected

3.  Search by Keyword:

o Allow search using natural-language or tag-based keywords (e.g., "white van", "man with helmet", "police uniform")

o Use object detection, pose estimation, and colour/attribute recognition

4.  Result Display:

o Show extracted frames along with timestamps and camera ID

o Provide options to export matching clips, reports, or annotated images

5. Frame Filtering:

o Fetch relevant frames only from long footages by keyword input or image input

Optional Features (Bonus)

• Multi-language support for keywords (e.g., Hindi, Gujarati)
• Filtering by time window, motion events, or activity zones
• Integration with facial recognition or license plate recognition APIs
• Report generation with chain-of-custody metadata
• Live-feed snapshot scanning (for future deployments)

Evaluation Criteria

• Accuracy and speed of frame extraction
• Robustness across different CCTV angles, lighting, and resolutions
• User interface simplicity for non-technical officers
• Innovation in combining vision and NLP technologies
• Scalability and offline usability

Bonus Points

• Real-time demo with sample CCTV footage
• Visualization dashboard (timeline view, search history, tag cloud)
• Use of lightweight models for field deployment
• Compliance with forensic standards (e.g., integrity hash for exports)