An automatic driving system for a car should be accurate and very fast in order to navigate a car safely. However, traditional system based on an embedded CPU cannot serve enough computing power to navigate a vehicle using vision processing in real-time. Interest point detection and matching which is the basis of many vision algorithms, such as object tracking, SLAM(Simultaneous Localization and Mapping) and object recognition, is hard to be run on typical embedded systems in real-time, because of the lack of computation resources.

The objective of this research is to propose a hardware accelerator through an in-depth analysis of these vision algorithms. We also focus on its scalability, low power consumption. It will be the foundation of a real-time vision processing platform for car navigation systems.