A Novel Solution for Camera Occlusion in Stereo Vision Technique

Abstract

Stereo version is a well-known technique for obtaining depth and surface information of objects from images captured by two cameras. In general, this technique is limited by camera occlusion inherent to grooved or steep profiles in 3D shape measurements. To alleviate this problem, a traditional dual-camera system accompanied with a projector based on the stereo vision principle is presented. The projector was considered as a virtual camera. Three sets of stereo vision systems were then obtained with each of the two cameras at the same global coordinates for 3D topography measurements. The data acquisition system constructed with the two real cameras was first calibrated. Then the two-camera-projector system can be calibrated based on the data obtained from the common measurement area of the three systems. Experiments were implemented to evaluate the accuracy of the merging profile integrated from the three sets of stereo vision systems. Experimental results show that the problem of camera occlusion encountered in the original dual-camera system was effectively resolved and a 0.05% error for the height measurement of a ladder-shaped specimen with camera occlusions was obtained in a measurement range of 640 mm × 480 mm.