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

In Phase I, we developed a suite of algorithms for the detection and classification of motorcycle and bicycle using IR LED stereo camera, LWIR thermal camera, and microphone. Test results using real data collected at streets have shown an excellent performance of this system. In Phase II, we will further improve the system developed and focus on the system performance issues in terms of near 100% detection/classification. We will also improve the algorithm computation speed to meet the near real-time requirements. In Phase II, we plan to use a low cost LWIR camera to make the product more affordable. To improve the classification accuracy, we will use an array of omni-directional microphones and develop algorithms to cancel out the random wind sound. Algorithms will also be developed to remove the motion blur in IR images and separate the sound signals from different types of motorcycles travelling in group. There will be two products to be developed in Phase II. They can be used at street intersections and street midblocks for the detection and classification of motorcycles and bicycles. Moreover, Phase II products are capable of interfacing with traffic controllers made by major manufacturers.