AI-Based Accident Severity Ranking and Emergency Response Automation System
Abstract: Road accidents require rapid assessment and coordinated communication with insurance and emergency services. This paper presents an AI-Based Accident Severity Ranking and Emergency Response Automation System that integrates image preprocessing, contextual accident-scene understanding, license-plate information extraction, deep-learning-based severity classification, severity-driven decision logic, and automated email notification. The system classifies an accident into Rank 1 (minor), Rank 2 (moderate or injury-related), or Rank 3 (severe) and maps the predicted rank to progressively broader response actions. The implementation uses a dataset of approximately 6,498 accident-related images with a 70:15:15 train-validation-test split. Phase 1 used MobileNetV2 as the baseline severity model, while Phase 2 introduced MobileNetV3 and a Gemini-3.6-Flash vision-language model for contextual scene description. The Phase 2 MobileNetV3 model achieved 93.35% accuracy, 94.26% precision, 93.35% recall, and a 93.80% F1-score. The integrated application combines the classifier output, accident-scene description, visible vehicle information, decision logic, a graphical user interface, and severity-specific email generation to provide an end-to-end accident-response workflow.
Keywords: accident severity classification, MobileNetV3, vision-language model, emergency response automation, license plate recognition, intelligent transportation systems
How to Cite?: Prithvi Nanjundan S, Dr. Pulkit Dwivedi, "AI-Based Accident Severity Ranking and Emergency Response Automation System", Volume 15 Issue 8, August 2026, International Journal of Science and Research (IJSR), Pages: 1332-1336, https://www.ijsr.net/getabstract.php?paperid=SR26817175856, DOI: https://dx.doi.org/10.21275/SR26817175856