Recording consistent multiview image data
Recording consistent multiview image data for 3D reconstruction is a challenging task, especially with low cost cameras. We propose a mobile setup of circularly arranged cameras for recording a rotational image sequence around the subject. It provides calibration of the cameras with a novel, multi-functional calibration target as well as software color adaption. The output of the calibrated cameras can be rectified to form a smooth rotational image sequence using the calibration data. We have proposed a complete setup for the acquisition of 3-dimensional image data and shown its application to the recording of a smooth image sequence with the viewpoint rotating around the subject. Samples of an image sequence recorded with the low-cost cameras are introduced. It should be noticeable that the subject is well-placed in the center in the rectified sequence (bottom) and that global color adaption has evened out most color discrepancies between the individual images. However, since the subject does not expose the same colors and illumination from every viewing angle, a global approach to color adaption is limited to minimizing an imprecise error function. The number of possible applications and extensions of this setup is manifold. Approaches to correspondence analysis are being tested at the moment and will offer new possibilities, e.g. for a correspondence-guided local color adaption. To reduce the number of cameras and generate new views, methods for image interpolation are applied to the data obtained from our setup. Different strategies for the camera setup will be researched as well, especially non-planar layouts are expected to yield advantages for 3D-based operations. A multi-view consistent segmentation of the subject from the background is also being researched at the moment.