Multi-channel edge detection

In vision-based spacecraft docking systems, scene structure is available over a range of different lighting conditions. In order to estimate the pose of the docking target, edges need to be extracted from this set of images. A multi-channel edge detection algorithm is developed to address this task that extends the single-channel Canny edge detector to operate on multiple channels. Input images are first processed in separate channels to obtain local gradient estimates. These local gradients are then modeled as a two-component mixture of Gaussian's in which the inliers (the normal gradient samples corresponding to the local edge structure) and the outliers (gradients corresponding to shadow edges and other random noise) are modeled by the two component Gaussians. The Expectation Maximization (EM) algorithm is used to decompose the mixture model, and to identify and separate the outliers from the inliers. A composite gradient map corresponding to the underlying edge structure is then recovered from the distributions of the inliers.; The composite gradient map computed by the multi-channel edge detection algorithm strengthens those parts of the image gradient that are consistent over the illumination changes associated with the input images, and discards the unstable parts. An edge map generated from the resulting gradient map better represents the underlying structure of the scene, since the influence of shadow edges, which are generally less stable under changes in illumination conditions, is reduced.