GPU and CPU Cooperative Accelerated Road Detection

In this paper, we propose a fast and robust unstructured road detection method that integrates GPU (Graphics Processing Unit) and CPU implementations. In order to ensure the robustness of the algorithm, BP (Back Propagation) Neural Network is employed to learn the color features from a set of sample of both road region and off-road region, and then to classify a newly pixel. And the B-spline curve model is employed to fit the boundaries of the lanes with the Least Square Method. To improve the real-time capability, the NVIDIA CUDA (Compute Unified Device Architecture) framework is used, and a GPU and CPU cooperative acceleration technique is proposed. Taking the advantages of these properties, the proposed implementation works out with high performance of detection in various environments. Meanwhile it is robust against noise, shadows and illumination variations. Moreover, it can performs about 10 times faster than a conventional implementation running on a CPU.