Photomotion

摘要

Traditional shape from shading techniques, using a single image, do not reconstruct accurate surfaces and have difficulty with shadow areas. Traditional shape from photometric stereo techniques have the disadvantage that they need all of the input images together at once to minimize the total cost, and this process must be restarted if new images become available. To overcome the shortcomings of the above two techniques, we introduce a new technique calledshape from photomotion. Shape from photomotion uses a series of 2-D Lambertian input images, generated by moving a light source around a scene, to recover the depth map. In each of the input images, the object in the scene remains at a fixed position and the only variable is the light source direction. The movement of the light source causes a change in the intensity of any given point in the image. The change in intensity is what enables us to recover the unknown parameter, the depth map, since it remains constant in each of the input images. This configuration is suitable for iterative refinement through the use of the extended Kalman filter. Our novel method for computing shape is a continuous form of the photometric stereo technique. It significantly differs from photometric stereo in the sense that the shape estimate will not only be computed for each light source orientation, but also gradually be refined by photomotion. Since the camera is fixed, the mapping between the depths at various light source locations is known; therefore, this method has an advantage over those which move the camera (egomotion) and keep the light source fixed. Results of this method are presented for sequences of synthetic and real images.