A Fourier-transform based method for autonomous path following

The problem addressed is that of finding the centre of a poorly defined path, for the purpose of allowing a vehicle to navigate the path autonomously. A cross-path profile is processed by some form of edge-enhancement, and thresholded. The poor definition of a semi-natural path means that applicable methods should not use small-scalelength detail, but rather should look at broad features and trends of the scene. The algorithm described arises from considering various possible `abstract´ properties of a scene. It is postulated that, if a vehicle has safely reached a given point, the future path would probably continue to be safe in the direction of maximum scene symmetry. The algorithm has been used to guide a small electric vehicle along a roughly painted track indoors. The fast-Fourier transform (of 32 pixels), with phase-unwrapping, took, less than 20 msec on a PDP LSI 11