Classification of Accident Risk Situations based on Vehicle Speed Estimation through Lane-Specific Coordinate Transformation
Seunghyun Kim, Hanbin Kang, Jonggyu Kang, and Changseok Bae
Journal of Korean Institute of Next Generation Computing, Feb 2025
This paper presents an accident-risk recognition model for intersections using CCTV footage. The model estimates vehicle speeds through lane-specific Bird’s Eye View transformations and classifies traffic situations into safe, caution, and danger levels via an MLP-based classifier. Experiments on real and simulated videos demonstrate 89% accuracy and 5 ms inference, validating its real-time applicability for accident risk detection.