Abstract: ALV （autonomous land vehicle） has a alluring prospect both in military and civilian applications. As the one of the kernel techniques in ALV, vision system has received close attention in the past two decades. In the paper here, a road recognition algorithm combining fusing of multiple information and matching with multiple models, in a knowledge\|based system KRUS is presented, with high performance achieved. In the algorithm, a unique edge detection method is developed, from which the benefit of good localization is received, while the disturbance of noise is reduced. Then the robust region information is fused with edge information, with the help of some knowledge of the scenes. The experiment results show that most of the ill\|formed or complicate\|structured images are very satisfying.