Abstract
<title>Abstract</title> <p>Road traffic accidents claim 1.35 million lives annually, with survival probabilities dropping 7–9% per minute of delayed intervention. Existing vision-based detection systems predominantly operate as isolated modules, suffering fromhigh false-positive rates and lacking integrated emergency response capabilities. This paper introduces InVAER, a novel AI framework unifying real-time accident detection with automated geospatial emergency coordination through three core innovations: (1) a multi-cue spatio-temporal collision validation algorithm integrating adaptive IoU thresholding, temporal persistence, velocity drop analysis, and trajectory convergence; (2) Adaptive Particle Swarm Optimization (APSO) for automated YOLOv8 hyperparameter tuning; and (3) a geospatial service identification module with multi-channel notification. Evaluated on 2,537 annotated frames, the APSO-optimized detector achieves 93.69% precision and 91.45% recall, while the full multi-cue framework reaches 95.88% precision and 95.20% mAP@0.5, processing at 30 FPS on edge devices. The multi-cue approach reduces false positives by 37%, and automated alerts reach emergency services in 4.2±0.8 seconds.</p>