7Q-04
Collision Detection and Responsibility Assignment in Traffic Accidents Using Object Detection, Knowledge System, and Open Data
○アベウノヘルトン ヤウヴ,菊地真人,大囿忠親(名工大)
This study addresses the timely determination of actor responsibilities in post-traffic accidents, essential for victim compensation. The previous system, utilizing driving recorder videos and a rule-based knowledge system, showed good performance but was limited to specific crash scenarios, requiring manual input. To overcome limitations, this research integrates open data, specifically leveraging the OpenStreetMap API, to enhance responsibility assessment. Improvements in crash and traffic light recognition models, along with OpenStreetMap API integration, significantly boost performance, especially in nighttime scenarios. The refined system achieves a 55% success rate, surpassing the previous 46%-53%, showcasing enhanced responsibility evaluation capabilities.