ABSTRACT
This chapter discusses various real-time challenges in the context of obstacle detection for autonomous driving such as small objects, occlusion, shadows, object size variations, and loss of depth information while projecting a three-dimensional scene on a two-dimensional plane. It is deep learning for obstacle detection, which concerns the selection of models and algorithms that can have their parameters tuned to improve the recommended performance criterion. The chapter considers the algorithms for obstacle avoidance using deep learning, different deep learning architectures that involve CNN and its variants, and the detailed YOLO algorithm for obstacle avoidance. It presents a huge range of research work on detecting and avoiding obstacles in the path of autonomous vehicles using deep learning frameworks. Datasets are important for scientists and designers, as a large number of the methods and apparatuses must be tried and tested before the autonomous vehicle is developed.
