A binocular stereo vision method is proposed for automatically locating the position and posture of workpieces, which is especially important when processing large, complicated structures, for example, the laser hardening and laser cladding of automotive die. First, a binocular stereo vision positioning system was designed and modeled, from which a method of background subtraction was proposed to extract the edge line of the foreground area. Furthermore, the intersection point of the workpiece contour line was taken as the characteristic point of the workpiece, and an algorithm that combines epipolar constraint with gray value similarity was proposed to quickly and accurately realize the feature points matching. Finally, experiments show that the workpiece can be positioned accurately, and that the precision of position recognition could be controlled within ±0.5 mm when the camera was 1 m away from the workpiece, meeting the requirement of robot processing.