Abstract
The growing of vehicle number brings about the needs of driver assistance system. For drivers to make the right decisions during the journey, it is important to make sure that drivers obtain all the information of surrounding environment and possible obstacles, especially vehicles. To achieve this goal, a growing number of papers dedicate in nding accurately vehicle detection. In this thesis, we propose a fusion vehicle detection system by combining the advantages of both radar and camera sensors. By applying radar data onto the vision image, an increasing accuracy of both position and contour is achieved. This system reduces the complexity and computing time in image processing for obstacle search. Here, a texture classied algorithm: Sorted Quadrant Median Vector (SQMV) is used to obtain the vertical and horizontal edge maps [1]. Based on these edge maps, we developed an algorithm to determine the existence of a vehicle in an image. The experiment results using MIT CBCL car database [2] demonstrate that this system can reach 98% detection rate with false positive rate 2.53%. In the last part of the thesis, a hardware model of SQMV edge detection is also implemented. The system can reach a frequency of 333 MHz, which means it takes only 50s to calculate the edge maps for a 128*128 image.