A Multi-Sensor Hybrid System for Vehicle Classification of Thirteen Classes

Image processing methods have been studied and implemented for the vehicle classification over the past decades. The most dominant approach is the use of vehicle dimension, which requires complex processes of sensor calibration as the vehicle dimensions shown in the imagery are heavily dependent on the camera’s field of view. In general, video systems have physical disadvantages related to the performance degradation under certain conditions such as occlusion, shadowing, inclement weather, sun glare and nighttime detection. In Phase II, we focused on the development of new hardware and detection algorithms, and developed three new products: MigmaMotorcycleTM, MigmaBicycleTM PB, and MigmaBicycleTM SP. Their field test results have shown that these products have the excellent performance. In this project, we will expand the core technologies developed in Phase II to the vehicle classification of all 13 classes defined by FHWA. The classification problem of 13 vehicle classes can be solved by developing a few algorithm modules using IR camera and microphones. Moreover, this system is easy to calibrate and low in cost. The vehicle classification results will also be sent to Amazon Cloud and users can view the vehicle classification labels and counts on the electronic maps.