EXPLORE PUBLICATIONS BY COUNTRIES

Edge Computing, Hardware Acceleration, Privacy-preserving, Real-time Crime Recognition, Smart Surveillance.

This paper presents an edge-based privacy-preserving in-vehicle device for real-time detection and reporting of weapons carried by attackers near a vehicle. It is no longer news that car owners are being attacked by criminals day by day, who often go unpunished. A lot of many lives have been lost, leaving car owners in a state of fear and uncertainty. Many have abandoned their cars trekking and camouflaging in order to escape from being attacked. This has raised a serious concern which led to the development of the proposed system. The system prototype integrates a motion sensor to trigger capture, a wide-dynamic-range camera for image/video acquisition, and a GPS module to record location metadata. All sensing and processing occur on-device: a fine-tuned YOLOv8 model runs on an embedded edge computer to detect and classify weapon types such as knife, gun, axe, face mask, machete etc. from images captured through vehicle windows. Detected events are logged in an encrypted circular buffer for deferred review and higher-accuracy offline processing; only anonymized event metadata (weapon type, confidence, timestamp, and hashed geo-ID) are transmitted to an authenticated online dashboard for immediate alerting. The system emphasizes privacy by design; ensuring raw footage is retained locally and released only under explicit authorization. Prototype evaluations demonstrate realtime performance with average end-to-end latency near 120–130 ms and mean detection precision exceeding 0.80 across target classes, while maintaining low storage and power overhead suitable for in-vehicle deployment.