A MVS based automatic 3D model reconstruction system from turntable image sequence

Traditionally, multi-view stereo (MVS) based 3D reconstruction needs a lot of manual interactions, such as camera calibration, silhouette extraction, texture synthesis, etc. To overcome this disadvantage, we present an automatic 3D reconstruction method based on image sequence. First, we design a new checkerboard with included angle 120°. The camera parameters and the rotational axis can be calibrated simultaneously. Then, an automatic silhouette extraction method is presented based on super pixel and cellular automata, from which the color image sequence can be segmented into binary images. Next, we use MVS based 3D reconstruction method to obtain 3D mesh model with camera parameters, color images and silhouettes. Finally, by using the 3D model, projection matrix and color images, we can synthesize the texture map. A prototype is developed based on Canon SDK and a controller. This system can help user to acquire 3D models without professional knowledge. Experimental results with variety of objects demonstrate the effectiveness of our system.