3D face and body modeling using camera and structured light

Three dimensional (3D) face and body models are needed in various fields such as plastic surgery, diagnosis of some diseases, tracking and monitoring of body growth and mutations, 3D computer games and 3D simulation software. Professional 3D face and body scanners are very expensive. Their prices increase according to the accuracy of 3D models and value of the brand. In this thesis, a low cost and high resolution 3D face and body modeling system has been developed. The developed algorithm and software generate 3D point cloud by using face and body images. The faces and bodies have been scanned and modeled by David SLS-2 optic scanner to collect reference data to evaluate obtainden results and for accuracy assessment. The intention of the developed software is to generate 3D model and achieve dense 3D points with high accuracy. In order to able to extract 3D point cloud from images, the imaging geometry has been determined according to algorithm. Therefore the face and body images have been taken through multi stereo geometry. A Canon 600D camera with 18 mega pixels resolution and 60 mm macro lens have been used for taking face and body images. Face images have been taken from the distance of 120 cm. The distances were nearly 650 cm when the body images have being taken. The developed software creates 3D point coordinates by using of 3456x5184 resolution image pairs. The output data of the software includes 3D points of face, body and other objects in the image pairs. Therefore the output data should be ordered to get only face and body models. The redundant areas of the output data are manually removed through using of Cloud Compare and Meshlab which are open source 3D point cloud processing softwares. The developed software creates 4,478,976 3D point coordinates by using of 3456x5184 resolution image pairs. The export data of the software includes some 3D points that are different objects from face and body in the images. Cloud Compare and Meshlab, open source point cloud processing softwares, have been used to remove unnecessary objects from point clouds. Five voluntaries have been modeled by the developed software and scanned by David SLS-2 to generate face models. Two voluntaries have been used for creating 3D body models with two different system systems. The face and body models have been obtained from both developed and David SLS-2 systems. Cloud Compare has used for accuracy assessment. The mean accuracy of the face modeling study has been calculated as %86.02. The mean accuracy value of the body modeling study has been %76.22. The accuracy values of the studies contain gross errrors which caused by moving of face or body, hair and brow. The thesis reveals a system that uses multi stereo images, a developed software and designed imaging geometry which are able to generate 3D point cloud from multi stereo images to reach 3D face and body models with low prices. The accuracy assessment of the developed system has been calculated by using of Cloud Compare. Keywords: 3D modeling, photogrammetry, point cloud, face modeling, body modeling