VisionGuard: Physics-Driven Vehicle Collision Detection and Alerting System using Deep Learning

Rajshekar Gaithond, Pallavi Jamadar — Wed, 24 Jun 2026 00:00:00 +0000

Traffic-related accidents represent a major source of global mortality, often compounded by delayed emergency assistance due to manual reporting constraints. This paper introduces VisionGuard, an automated, web-based collision detection and alerting platform that integrates deep learning-based object detection, multi-object tracking, and physics-driven kinematic reasoning. Operating on live or pre-recorded traffic video streams, the system utilizes the YOLO v11 nano model to localize vehicles across four categories (cars, motorcycles, buses, and trucks) and maps them to a ByteTrack tracker to maintain unique identities across successive frames. A custom physics engine records center-point trajectories in a sliding window buffer, continuously evaluating vehicle pairs for spatial proximity (Intersection over Union and center-point Euclidean distance) and kinematic anomalies. A collision is logged when close proximity coincides with a kinetic shock (sudden velocity drop exceeding 70 %) or an abrupt angular direction change. Confirmed incidents trigger visual overlays, capture localized snapshots, and generate structured, evidence-backed PDF reports. Experimental results on real-world CCTV footage show that VisionGuard achieves a collision detection accuracy of 95.05 %, a specificity of 97.36 %, and a processing speed of 45.5 FPS under GPU acceleration, providing an explainable and reliable automated traffic monitoring layer.

IoT-Enabled AI System for Real-Time Fake Currency Detection and Denomination Identification

N. B. Mahesh Kumar, Bharath J, Arun Sanjay A, Dhanvanth Priyan S, Aswin G — Mon, 01 Jun 2026 00:00:00 +0000

The proposed system proposes a low-cost, sensor-driven embedded system using an Arduino microcontroller to address the rising threat of counterfeit currency. Unlike resource-heavy machine learning approaches, this architecture utilizes light sensors—such as a Light Dependent Resistor (LDR) or BH1750—to analyze the optical characteristics of banknotes. The system works by illuminating a note with a controlled light source and measuring the reflected or transmitted intensity. Because genuine currency possesses unique materials and security features, it produces consistent optical patterns that differ significantly from counterfeits. These readings are processed via a threshold-based algorithm to verify authenticity instantly. Beyond detection, the system features automatic denomination recognition and counting. Since each note value corresponds to a specific light intensity range, the device identifies the denomination and updates a cumulative total. Results, including authenticity status and total value, are displayed on an LCD module. Designed for portability and energy efficiency, this solution is ideal for small businesses, retail shops, and banks. Experimental evaluations show high accuracy and fast response times, effectively reducing human error in financial transactions. While currently optimized for visible light, future iterations may integrate UV or IR sensing to further enhance robustness against sophisticated forgeries. This approach offers a simple, accessible, and effective alternative to complex, expensive detection hardware.

Journal of Sensor and Cloud Computing (e-ISSN: 3048-9199)

VisionGuard: Physics-Driven Vehicle Collision Detection and Alerting System using Deep Learning

IoT-Enabled AI System for Real-Time Fake Currency Detection and Denomination Identification