Abstract
This chapter addresses the challenging problem of detecting and classifying speed limit signs in a real-time video stream using an embedded, low-end GPU. Graphics processing units (GPUs) have been increasingly used for applications beyond traditional graphics that are well suited for their capabilities. One of these fields is automotive computing. GPU-based techniques can be used for performing real-time speed-limit sign recognition on a resource-constrained system with a low-end GPU that can be embedded in a car. Three pipelines are implemented to address this problem. The first is a detection-only feature-based method that finds objects with radial symmetry. The second is a template-based method that searches for image templates in the frequency domain using Fast Fourier Transform (FFT) correlations, suitable for both EU and US speed-limit signs. The third is the classic GPU-based SIFT approach that provides a basis for evaluation of recognition results of the template-based approach. To provide fast runtimes and make efficient use of the underlying limited hardware resources, the inherent parallelism in the recognition process was exploited using data-parallel algorithms that are suitable for the GPU architecture. This study serves as proof of concept for the use of GPU computing in automotive tasks. However, in order to make the best use of an embedded GPU in the cars, it should be able to simultaneously run multiple other automotive tasks that are a good fit for the GPU architecture.