Acquisition of 3D-Head-Models using SLR-Cameras and RGBZ-Sensors
For automatic authentication of subjects at access controls or security gates it can be useful to have a three-dimensional model of a subject's head, for example to improve the performance of face-recognition algorithms by adjusting lighting or pose. This thesis describes a system for the automatic acquisition of a well textured face model of a subject walking through a security gate. Two Kinects and an SLR-camera are used to gather depth-data and to take a high quality picture of the subject's head. The collected depth-data is merged to obtain an improved head model. Since the subject is moving, all depth-scans must be aligned before further processing can be performed. For the alignment process it's important to consider that head and torso are not tied rigidly together, so only the head alone can be used to perform the alignment.
By employing the ICP-algorithm and warp-optimization, the collected depth-scans are aligned to each other and to the SLR-image. Then, the depth scans can be merged to obtain an improved facial geometry and the SLR-image can be used to replace the low definition Kinect-image. The main contribution of this work is the registration of consecutive depth scans from both Kinects and the alignment of the resulting model to the SLR-image.
Raúl Rojas, Prof. Dr.-Ing. Peter Eisert; David C. Blumenthal-Barby (Supervisor)
Master of Science (M.Sc.)