I. Introduction
Nowadays, autonomous vehicles and advanced driver-assistive systems (ADAS) perception problems like obstacle detection and lane detection are among the hot areas in computer vision. Ultimately, in each task, the target is to reach a sufficient understanding of the scene around the vehicle that is enough to make safe and efficient control decisions on behalf of the driver. What differentiates autonomous vehicles perception problems from other computer vision problems is the required method of quality in terms of accuracy and speed. From one side, safety measures demand highly accurate algorithms to ensure reliability, and from the other side, the severely limited resources on vehicle-based systems demand computationally inexpensive algorithms.