This paper describes a vision global location system for mobile robot systems. It uses a global observation sensor to observe some artificial landmarks mounted in a work place. And then a data processing is used to obtain the sensor position and orientation in the environment. This paper mainly describes some improved methods of reliability and image processing speed for image target recognition, the optimal method for location calculation is mentioned as well. At last, the experiments are given to testify that such methods used for the vision sensor system are effective to guide a mobile robot system by global environments.