Automated Vehicle Collision Detection: Multi-Model Computer Vision Framework

Road traffic accidents account for more than 1.35 million deaths each year, imposing an economic burden of roughly three percent of global gross domestic product (GDP) [1]. Conventional sensor-based accident-detection systems suffer from high deployment costs and limited contextual awareness. This paper presents a comprehensive multi-modal framework that fuses a six-model YOLO ensemble with BLIP image captioning to detect and describe vehicular collisions in real time. Extensive experiments on heterogeneous datasets demonstrate a mean detection accuracy of 94% at sub-second latency, outper-forming state-of-the-art baselines while remaining feasible for smart-city edge deployment. Our framework processes video streams at 10 FPS while maintaining high accuracy across diverse environmental conditions including nighttime, adverse weather, and varying camera angles. The integration of natu-ral language descriptions provides emergency responders with actionable intelligence, reducing response times by an average of 15%. Implications for emergency response, urban traffic management, and future intelligent transportation systems are discussed. Additionally, the proposed framework demonstrates improved interpretability through natural language captions and efficient deployment scalability on edge AI devices, making it a promising step toward autonomous, privacy-aware traffic safety systems.