Corner detection and matching for visual tracking during power line inspection

Power line inspection from a helicopter using video surveillance techniques demands that the camera be automatically pointed at the object of interest, in order to compensate for the helicopter's movement. The possibility of using corner detection and matching to maintain the fixation point in the image is investigated here. An attractive feature of corner-based methods is that they are invariant to camera focal length, which can vary widely during inspection. The paper considers the selection, parameter determination and testing of a customised method for detecting corners present in images of pole-tops. The method selected uses gradient computation and dissimilarity measures evaluated along the gradient to find clusters of corners, which are then aggregated to individual representative points. Results are presented for its detection and error rates. The stability of the corner detector in conjunction with a basic corner matcher is evaluated on image sequences produced on a laboratory test rig. Examples of its response to background clutter and change of illumination are given. Overall the results support the use of corners as robust, stable beacons suitable for use in this application. (C) 2003 Elsevier B.V. All rights reserved.