Abstract: A few techniques are applied for occlusion handling in video-based augmented reality.Epipolar geometry constraint is calculated by nonlinear error propagation at initial time to accelerate the speed of stereo matching；global and local methods are integrated to extract foreground objects precisely；and only pixels on objects’contour are reprojected into 3D space to fasten the depth calculation process.In comparison with other algorithms,our algorithm has advantages of efficient computation,high accuracy and easy implementation.Experimental results show the correctness and feasibility of the algorithm.