Abstract

Public-space surveillance produces large volumes of video that can be difficult for personnel to review promptly. This paper presents Spot Crime, a prototype surveillance-support system that combines face detection, CNN-based embedded facial feature representation, database-based similarity matching, and alert generation in a web application. The proposed pipeline accepts CCTV frames or uploaded images, preprocesses the input, detects and aligns facial regions, generates facial embeddings, and compares them with authorized reference records. When a similarity score exceeds a threshold validated for the selected model and operating conditions, the system creates a preliminary alert containing the timestamp, camera identifier, candidate score, and reference-record identifier for review by authorized personnel. OpenCV and dlib support image processing and face localization, while Python/PHP and MySQL provide application and record-management functions. Unlike a system that labels a person as a criminal, the proposed design treats a match only as a candidate identity association; it does not establish guilt, and human verification remains essential. The paper details the architecture, workflow, database structure, operational safeguards, and an evaluation framework. Since quantitative benchmark results are not provided in the current manuscript, claims about accuracy, latency, and field effectiveness are treated as objectives for future validation rather than established outcomes. The principal contribution is an integrated design for traceable candidate matching and alert review, with explicit attention to false matches, demographic performance, privacy, and responsible operational use.

Authors

A. Nandhini, M. Amilesh, P. Archana, A. Nisath, R. Siva
Nehru College of Management, India

Keywords

Public-Space Surveillance, Video, Spot Crime, CNN-based Embedded Facial Features