Segmentation of forward looking images of suburban roads using some simple texture measures is described. Driveable regions are labelled using a Markov random field model. High resolution is achieved by a double pass method. Considerable interest is being shown in developing the capability to drive automated vehicles through road systems and more complex environments typical of industrial installations. Recent work has tended to concentrate on the techniques necessary to process, in real time, TV images obtained from cameras mounted on such a vehicle. Suitable processing of the images provides an input to, principally, the heading control system. The approach has the advantage of flexibility and maybe cheapness when compared with systems using markers, painted lines or buried cables. Because of the potentially large amounts of data which would need to be processed in real time, the reported work has proposed solutions to the relatively simple problem of driving along well defined, uncluttered roads [1, 2]. For several reasons, it is not evident that the methods which have been described would be applicable to complex or ill defined road systems and open terrain. In the work described in this note, an attempt has been made to segment scenes on the basis of texture. Texture features are not necessarily easy or quick to calculate. However they do have the advantage of allowing simple, closed regions to be extracted. The procedure is essentially simple. A set of features is defined and statistically significant numbers of these are measured for the types of texture that it is required to identify. Feature values for a sample of the unknown texture are then compared with those of the reference set. A label is attached to the sample according to the closeness of the two sets of features. The labelling process should take into account the probability that several textures may produce similar statistics and that the matching may not be exact. No more than three features have been used in the current work, a Markov process being used to produce final, labelled images of suburban and country scenes obtained from a forward looking camera.

Some data obtained from the simulation of a multicomputer system are described and compared with results from a semiempirical analysis of the system. The analysis indicates the dependence of system efficiency on the method of scheduling a job and on the job characterization parameters.