Enhancing Safety and Collision Avoidance in Autonomous Vehicles through Pose Estimation Techniques

Autonomous cars are rapidly becoming essential and widely adopted due to the convenience and improved quality of life they offer. Pose estimation is a critical component in autonomous driving systems, enabling vehicles to accurately perceive, understand, and navigate their complex surroundings. By determining the precise orientation and position of surrounding vehicles, autonomous systems can make informed decisions for collision avoidance, lane-keeping, and efficient traffic management. This paper proposes a novel approach to vehicle pose estimation using the YOLOv8 deep learning model. The model was trained with a comprehensive dataset, leveraging its advanced pose estimation capabilities to achieve high accuracy and real-time performance. Before training, hyperparameter tuning was carried out with the dataset to optimize the model’s performance. To further improve the model’s effectiveness, a custom dataset is being proposed which is captured from the unique traffic condition of Bangladesh roads, aiming to enhance vehicle pose predictions in the context of Bangladeshi traffic scenario. Our experiments demonstrate that YOLOv8 significantly outperforms traditional methods in terms of both accuracy and computational efficiency. The quantitative and qualitative results of the predicted key points, highlight the vast potential of YOLOv8 for enhancing autonomous driving systems, providing a robust solution for real-time car pose estimation in diverse and dynamic traffic scenarios.