Review of Dynamic Obstacle Avoidance for Autonomous Driving Based on Binocular Vision

Abstract

With the rapid development of automatic driving technology, the vehicle's ability to accurately perceive and quickly respond to the surrounding environment has become the key to realize safe and efficient driving. Within this landscape, binocular vision technology has distinguished itself as a game-changer. Its capacity to capture real-time, three-dimensional environmental data offers preeminent advantages in tackling one of autonomous driving's most critical challenges: dynamic obstacle avoidance. This paper begins by exploring the application of binocular vision in Mars rovers, illustrating its pivotal role in autonomous obstacle avoidance. Building on this foundation, this paper then delves into the fundamental principles of binocular vision, including depth calculation methods and the overall system architecture. Then, this paper discusses in depth the dynamic obstacle avoidance strategies in autonomous driving, which are closely dependent on the real-time and accurate environmental information provided by the binocular vision system. In the concluding sections, this paper shines a spotlight on the primary hurdles confronting current dynamic obstacle avoidance technologies. These challenges range from adapting to monotonous environments like arid desert, to pushing the boundaries of hardware capabilities, and refining algorithms to achieve split-second decision-making. This paper provides valuable references and lessons for the practical application of automatic driving perception technology.