Because of a recent federal initiative, states are now required (as of June 2008) to collect and submit motorcycle VMT data to the FHWA. These dataare needed to obtain better counts of motorcycles to evaluate their impact on crashes and traffic flow. However, there is concern about the quality of data submitted. Many states have identified problems with using automatic traffic recorders to account for motorcycle traffic. Existing sensors exhibit difficulties in counting motorcycles that travel side by side or close behind each other, they have difficulty in distinguishing larger motorcycles from passenger vehicles, and magnetic counters in particular donot sense motorcycles that do not pass over or travel close enough to thesensor. Alternatively, some states conduct manual classification counts, but these efforts are labor intensive and lead to sparse data. A further complication is that classification counts are frequently conducted during the week and therefore do not capture weekend motorcycle traffic numbers.This paper evaluates a video based traffic monitoring system developed atClemson University that is capable of classifying vehicles including motorcycles. The processor uses vehicle tracking rather than virtual detection as a means to collect vehicle count, speed, and classification data. Motorcycles are classified using an algorithm that calculates a vehicle's length, width, and height through a series of frames. The system is evaluated using traffic data containing more than 2000 motorcycles collected at two locations in Myrtle Beach, South Carolina during a motorcycle rally. Thedifference between actual and system motorcycle counts ranged from 0.6% to just over 6% depending on direction and location. The difference for all vehicles ranged from 0.25% to 3.6%. The system successfully classifies motorcycles traveling in close pairs and in small groups, while it experiences difficulty in cases of severe occlusion.
Abstract