A New Multi-Sensor Hybrid System for Motorcycle Detection and Classification

Detection, classification, and characterization are the key to enhancing motorcycle safety, operations and travel estimation. Motorcycle fatalities are currently estimated at 30 times those of auto fatalities per Vehicle Mile Traveled (VMT). Although it has been an active research area for many years, motorcycle detection and classification still remains as a challenging task. The goal of Phase I development is to create advanced technologies to detect motorcycles, classify them separately and accruately from other vehicles, and identify different kinds of motorcycles such as heavey touring-class motorcycles, light motorcycles, motor scooters, mopeds, bicycles and tricycles. Under our previous FHWA SBIR Phae I/II funding, we have developed an IR LED stereo-vision based system for robust pedestrian detections. In Phase I, we will modify and enhance our existing pedestrian detection algorithms to detect motorcyclists, and consequently, the motorcycles. We also propose to add IR thermal cameral which can capture some unique thermal features associated with various types of motorcycles. To further enhance classification performance, we propose to use acoustic sensors to classify motorcycles from their unique sould profiles. Detections from these hybrid sensors will be fused to achieve a refined detection. Field tests are planned in Phase I to evaluate the system performance.