<![CDATA[Autonomous vehicles (AVs) are an emerging innovation in intelligent transportation systems, yet traffic accidents remain a critical concern due to environmental uncertainty and sensor limitations. This study aims to analyze collision risk levels in autonomous vehicles using a Bayesian Convolutional Neural Network (Bayesian CNN) integrated with the Monte Carlo Dropout (MC Dropout) technique. The model was trained on 11,000 visual datasets from the Central Bureau of Statistics (BPS) and synthetic data representing diverse road conditions. The Bayesian inference framework enables dynamic and adaptive risk prediction by continuously updating posterior probabilities based on sensor input changes. Simulation experiments were conducted using a Python-based interactive interface (pygame) to visualize vehicle movements and real-time collision probabilities. Results show that 48% of test scenarios were classified as very low risk (0–10%), 28% as low (11–30%), 16% as medium (31–60%), and 8% as high (61–80%). The model achieved a reduction in loss value from 0.43 to 0.08 and maintained 76% of simulations within low and very low risk categories, confirming system stability and reliable convergence. The findings demonstrate that the Bayesian CNN model effectively captures uncertainty and provides adaptive, probabilistic predictions, supporting safer and more intelligent autonomous vehicle operations.]]>
Copyrights © 2025
