Deep Vision for Navigation of Autonomous Motorcycle in Urban and Semi-Urban Environments

Deep neural networks are currently the best solution for road and traffic scene interpretation for autonomous and self-driving vehicles. Compared to the autonomous cars, motorcycles have significant flexibility and advantages in crowded traffic situations and especially in non-urban and off-road areas. Many off-road tracks especially for agriculture and environment management tasks are only traversable with motorcycles. In this paper, a deep neural network is used for design and implementation of the vision system for navigation of an autonomous motorcycle. The proposed framework is evaluated using real world scenarios captured by a real motorcycle in various complex situations. The experimental results show that the proposed framework is capable of highly accurate interpretation of various environments for autonomous navigation of a motorcycle.