Road Traffic Sign Detection And Recognition

Road vehicle guidance system based on machine vision has three main parts: 1) road detection, 2) identify obstacles, 3) sign recognition. The first two have been studied for many years, and have achieved very good results, but the traffic sign recognition is a very little involved in the field. Traffic signs can provide drivers with the information in front of the road, so that drivers can be more convenient and more secure when driving, so the traffic signs play an important role in vehicle driving. In order to distinguish them from the natural environment, they are designed to be easy to identify. The algorithms described in this paper mainly use these features, which are mainly two parts. The first part is the detection, using color threshold segmentation image and shape analysis to detect the signs; the second part is the identification and classification, the use of neural network.Common traffic signs warning, prohibitions, instructions, the color and shape of these three kinds of traffic signs are very easy to detect. In this paper, the color detection method for RGB color space, it is characterized by intuitive, easy to understand, high real-time. But there are also shortcomings. It is particularly sensitive to light, which requires us to light conditions in the case of a good collection of pictures. Because the corner is one of the important features of the shape, so this paper uses the best corner detection traffic signs, this algorithm constructs different masks for different corners, detection of corner by convolution operation of mask and image, then, you can know what’s on the corner by the corner. The neural network is used in recognition and classification. From the point of view of module identification, the biggest advantage of the neural network is its ability to track and identify nonlinear characteristics and for the identification of traffic signs picture has a very good effect.In this paper, the detection and recognition of traffic signs in the process and algorithms in many places need to be improved and further understanding of shape detection and neural networks is also required.