Real-time Traffic Sign Recognition with Map Fusion on Multicore/Many-core Architectures

This paper presents a parallel implementation and performance analysis of a system for traffic sign recognition with digital map fusion on emerging multicore processors and graphics processing units (GPU). The system employs a particle filter based localization and map matching and template-based matching for sign recognition. In the proposed system, a GPS, odometer and camera are fused with digital map information. The system utilizes the depth sensor of a Kinect camera for the detection of signs and achieves high recognition rates for both day and night conditions. Tests were performed on real data captured in the vehicle environment comprising various road and lighting conditions. Test results show that speed increases of up to 75 times for localization and 35 times for sign recognition can be achieved on parallel GPU implementation over sequential counterparts. As those speedups comply with real-time performance requirements, high computational cost of using map topology information with large number of particles in localization implementation and template based matching for sign recognition is proven to be handled by emerging technologies. The system is unique since it is not limited to certain sign types; it can be used in both day and night conditions and utilizes a Kinect sensor to achieve a good price/performance.