Study On Traffic Sign Recognition Based On Machine Learning

With the developing economics and growing urbanization, the growing phenomena of the traffic congestion and frequent traffic accidents have a strong impact on the development of modern city. In response, intelligent transportation system (ITS) emerges and develops rapidly. As one of the main development directions of intelligent transportation system, research on traffic sign recognition has become more and more important. However, with the road environment becomes more and more complex and diverse, the challenges and difficulties on traffic sign recognition are rising. Study on the traffic sign recognition methods, which can satisfy real time and veracity requirement of traffic sign recognition in real road environment, has great signification for the development of intelligent transportation system and the construction of a smart city. Under this background, aims to promote the accuracy and efficiency of the traffic sign recognition algorithms, this dissertation has deeply studied the traffic sign recognition methods in theory and practice, and the research contents and innovation are as follows:(1)In view of classification and design features of Chinese traffic signs, original images under different environmental conditions are acquired by using digital camera and from screenshot based on Baidu panoramic map, which include multi-class traffic signs, two traffic sign image databases are built which contain more than60kinds of common traffic signs, in total of2401images and provide image resources for the subsequent research work.(2)Considering the illumination normalization and filtering existing in traffic sign image preprocessing, Gamma correction and filtering algorithm based on4-neighborhood operator be given to improve the quality of the original image. Then studied and analysis the traffic sign image based on color and shape features, a new method which the detection and segmentation of traffic signs are completed by using RGB color space and Canny edge detection operator been offered, and the effectiveness of the method is demonstrated by experiment.(3) Firstly, traffic sign images are roughly classified into six sub categories based on color and shape feature; then for every sub category, further fine classification is completed by using machine learning algorithm. Consequently, two new methods for fine classification based on machine learning are proposed, the first method is based on principal component analysis-support vector machine (PCA-SVM) with grid search (GS), the other method is based on histogram of oriented gradient-probabilistic neural networks (HOG-PNN) with grid search (GS). The performance of these two methods is analyzed under the same experimental environment, and effectiveness of the proposed method is demonstrated by a great deal of simulations and experimental.In this dissertation, there are40illustrations,15tables and69references.