What Car is That ?

Napier New Zealand July 2001 www.naccq.ac.nz INTRODUCTION The issues and problems of vehicle recognition are those of all forms of computer-based object recognition studied over the last few decades. Initial promising progress has given way to a complex of sub-problems and a range of heuristic methods. A human expert may be able to quote the make and model of a car given only a glimpse of it on a highway, yet the problems presented to a computer expert given the same images, are not as easily solvable. If our task is to identify each vehicle in a stream of traffic according to make and model, then we have to confront problems of tracking, occlusion, colour, shading, pose and template comparisons. While there has been much research into each of these fields in the general object identification area, there seems to be no attempt to solve the automatic recognition of vehicle by make and model in a moving stream of traffic. Some sub-problems involving the same scene have been solved. We will identify work on some of the sub-problems such as number-plate identification and class-of-vehicle discernment, as well as give an account of a plan to constrain the environment so that careful metric data can be extracted from an image, thereby leading to a look-up table of car lengths, heights and similar measurements. ABSTRACT