Multi-Model Deep Learning Framework for Autonomous Perception and Scene Understanding

Autonomous perception and scene understanding is a crucial task when it comes to groundbreaking inventions such as autonomous vehicles (AVs). AVs are capable of understanding their surroundings and determining what to do on their own. However, an essential aspect of ensuring their reliability is understanding the certainty of their predictions. This paper addresses the above by introducing a probabilistic approach that can estimate the model’s certainty in its predictions. Additionally, it proposes a new model specifically designed to evaluate prediction certainty. It also presents a comparative study of many state-of-the-art approaches aimed at improving autonomous perception technologies.