Share
Export Citation
Stereo Vision Based Vehicle Detection and Distance Estimation for Braking Anomaly Detection
Rahmat M.A.
Proceedings 7th International Conference on Informatics Multimedia Cyber and Information System Icimcis 2025
Abstract
Detecting sudden braking anomalies is crucial for enhancing traffic safety, particularly in dense urban environments such as Makassar, Indonesia. This study proposes a stereo camera based framework combined with preprocessing techniques to improve the accuracy of vehicle detection and distance estimation. A total of 3,270 stereo images were used to train and evaluate multiple YOLO models (YOLOv5 to YOLOv8) under two scenarios: one without preprocessing, and another with an enhanced preprocessing pipeline. Experiments were conducted over 100+ epochs using an NVIDIA Tesla T4 GPU. In Scenario 1, YOLOv5 achieved a mean average precision (mAP 50) of 92.90% and a mean absolute percentage error (MAPE) of 8.40%. In Scenario 2, after applying data augmentation, normalization, and contrast enhancement, the mAP 50 improved to 97.50% while the MAPE dropped significantly to 2.11%. These results demonstrate that integrating stereo vision with robust input enhancement can significantly boost object recognition and distance estimation, laying a solid foundation for effective detection of sudden braking events in complex traffic settings.