New algorithms for contour-based processing and interpretation of images

The paper reports development of a simple and reliable multi-comer detection technique that combines local computational schemes and an inexpensive model-based matching approach. The input data are digital contour images extracted from camera-captured scenes by using any edge detector. Although most of the results have been obtained by using the software implementation, the algorithmes structure has been optimised for hardware implementation. A preliminary hardware design has been proposed, and some of the modules have been already implemented in FPGA. Eventually, it should be possible to detect on-line all multi-comers (and to determine their geometry) within images incoming at standard (e.g. TV) rates. The algorithm has been developed within a joint research project of Robotics Research Centre (NTU) and British Gas Asia Pacific on development of an autonomous underwater vehicle for specialised applications