Vision-Based Vehicle Wheel Detector And Axle Counter

This paper introduces a method for counting vehicle axles in the video sequence appropriate for use in vision-based automatic vehicle classifiers with narrow field of vision. The procedure is simple, robust to wheel detector errors, and computationally efficient due to the use of the fast Hough transform. It provides a self-test for applicability for detection of cases where it is not applicable, thus allowing adaptive method switching for cases of non-uniform speed. We also consider extensions to the method allowing for systematic false positive wheel detections and non-uniform motion.