A Vision System For Intelligent Vehicle Using DSP

Since traffic environment heavily depends on visual information, machine vision is a must in iTS. Image processing plays an important role in ITS. This dissertation presents a real-time vision system for intelligent vehicles. There are three major components in the system. They are imaging (image input), and software (algorithm), processing hardware. First of all, the video images acquired by a forward looking video camera install on a moving vehicle on freeway. Various traffic conditions and visibility conditions are included in the images sequences. Second, lane and moving object detection and tracking are two major parts in the algorithms. And their features are extracted in a small area estimate by The Kalman filter algorithm based on past observations. And the square root algorithm is used in the filter to make the estimation more accurate and stable than the conventional non-square toot algorithms. For robust recognition in various conditions, we propose three methods for lane detection; a robust white line detection using deformable template approach, an accurate line direction calculated using an equation, and an integration method of them. The algorithms can smoothly track white line through maneuvers such as lane changes. For recognition various objects, we propose a method which only detect angle points instead of contour .The method are robust to reduced visibility conditions. All the methods can be used for vision-based driving assistance and autonomous vehicle. Last, the hardware system consists of a PC Pentium 266 Mhz with MMX technology and a frame-grabber board embedded a C620 1 .The frame rate is 25. The algorithms have been experimented in real-time and the result is good. In conclusion,it is a simple and feasible Intelligent vehicle vision system. Liu Zhong.....yun Directed by Su Kai_na...