基于CUDA的交通限速牌识别

Speed Limit Traffic Sign Recognition Based on CUDA

智能识别交通限速牌对智能驾驶车辆速度调节和安全行驶有着重要意义。为有效解决交通限速牌识别问题,提出一种基于统一计算架构(Computer Unified Device Architecture,CUDA)的实时交通限速牌识别算法。利用HSV颜色空间提取ROI(Region Of Interest)区域,结合霍夫圆算法准确定位交通限速牌位置,设计一种基于卷积神经网络(Convolutional Neural Network,CNN)的算法模型对定位区域进行数字识别。使用NVIDIA GTX680显卡对算法进行加速,可达30ms/帧的处理速率,基本满足了实时性要求。实验证明,基于CUDA的交通限速牌识别算法具有良好的适应性和准确性。

In the intelligent driving system,the intelligent identification of speed limit traffic sign has very important significance for the vehicle speed adjustment and driver traffic safety.To effectively solve the problem of speed limit traffic sign recognition in real scenarios,a real-time speed limit traffic sign algorithm based on computer unified device architecture（CUDA）is proposed.We use HSV color space to extract region of interest（ROI）area and accurately locate the position of speed limit traffic sign using Hough circle algorithm;and then an algorithm model based on convolutional neural network（CNN）is designed to recognize numbers in the positioning area.NVIDIA GeForce GTX680 GPU（Graphic Processing Unit）is used to accelerate the algorithm on Compute Unified Device Architecture（CUDA）platform and achieve the processing speed of 30 ms per frame which basically meets the real-time processing demand.Experiments show that the CUDA-based speed limit traffic sign recognition algorithm has good adaptability and accuracy.