Large-scale three-dimensional vision measurement based on stereo vision is an effective and flexible method for precise space coordinates measurement when the object′s shape is irregular and the size is relatively large. Different from traditional methods, the accuracy of large-scale vision measurement degrades with distance rapidly. Meanwhile, it is hard to control the illumination under large-scale condition, which also makes the measurement quite unsteadily. Thus how to maintain the accuracy under large-scale situation becomes the key technique problem to be solved. The principle of the vision measurement system is introduced. The factors which affect the measuring accuracy are analyzed. Structured light extraction method with phase congruency and epipolar constraint is proposed which can increase the image analysis accuracy obviously. A prototype is developed and the experimental results show that the proposed method is of high accuracy and large measuring range.