Image Processing-Based Vehicle Class Identification in Mixed Traffic

Abstract

Traffic composition plays a major role to characterize a heterogeneous traffic stream where various categories of vehicles with diverse static and dynamic characteristics share the common carriageway. Due to this diversity, the manual collection of classified traffic volumes often becomes difficult, particularly on busy roads. In this regard, the popularity of the videography method has increased significantly in the last few decades. Because the method offers flexibility to the enumerator(s) to extract the required data at their convenience by playing the video file on a computer screen. However, this method demands ample time and effort from the enumerator(s) that is considered as its major drawback. On this background, the present study proposes an advanced image processing-based approach which is helpful to identify the vehicle class in a video exhibiting the mixed traffic movements. The proposed methodology addresses two major limitations associated with the existing approaches; i) the problem of varying blob size, and ii) the thresholding problem. By eliminating these limitations, the proposed methodology has yielded an accuracy of 96.82% in identifying the right vehicle class. Apart from this, the proposed approach will significantly reduce the time and efforts devoted by the enumerator(s), and hence, it can be a decent substitute for the manual extraction of the classified traffic volume in the future.