Fuzzy C-Means Image Segmentation Approach for Axle-Based Vehicle Classification

Abstract

Vehicle classification information is vital to almost all types of transportation engineering and management applications, such as pavement design, signal timing, and safety. Although the vehicular length–based classification scheme is widely used by state departments of transportation, this scheme lacks the capability of accurately producing axle-based classification data. Limited by the capital cost, axle-based vehicle classification data sources are very narrow. This paper presents an image segmentation–based vehicle classification system with an attempt to increase the efficiency of axle-based vehicle classification. The video-based vehicle classification system Rapid Video-Based Vehicle Identification System (RVIS) is developed to identify the number of axles automatically from ground-truth videos. Through the testing of individual vehicle image data sets, it is shown that the RVIS system is capable of successfully detecting all FHWA 13 vehicle classes. However, larger-scale testing of the RVIS system with a predetermined set of morphological parameters produces less accurate results. Comparison of two testing hours shows that with greater effort in calibration, results can be improved significantly and a great potential for field application exists. The advantages of the RVIS system are its robust and fast algorithm and its flexibility in that it can be applied either from a mobile video source or at locations with traffic-monitoring videos available. The RVIS system is a proven vehicle classification data source that adds to other existing vehicle classification approaches.