Research Of Fast And Robust Traffic Sign Detection And Recognition In Complicated Environment

Detection and Recognition of traffic signs have been a hot research area in the interdisciplinary field of intelligent transportation and machine vision, and are closely related to urban transport planning and traffic safety, which has very important theoretical and practical value. Over the last decade, the sign detection and recognition based on machine vision technology have made great progress, and many excellent algorithms appeared. However, the current sign recognition system still faces many technical problems to be solved. In this paper, some key points in the research of traffic sign recognition system are to be research, such as the extraction of the regions of interest, traffic sign detection and traffic sign recognition. We also research the traffic sign detection and recognition methods with big occlusions or with large shooting angles. The main contents and innovations are as follows:(1) A ROI extraction method based on the characteristics of high contrastThe traditional methods of extracting the region of interest (ROI) based on color threshold are sensitive to color changes. This paper proposes a method of extracting regions of interest based on the characteristics of high contrast, called High Contrast Region Extraction (HCRE). The principle of designing this method is that different kinds of traffic signs have significant differences in the distribution of its sign area; whereas the local area of the surrounding environment usually has similar distributions, such as roads, sky and trees. First, the input image is color-enhanced; then, the high-contrast regions can be extracted using the cumulative voting of high contrast features; finally, according to the voting results, the ROI regions can be extracted. This method is more robust to color changes and fuzzy edges. Based on experiments on the sign databases from three different countries demonstrate the effectiveness of the proposed ROI extraction method.(2) Traffic sign detection based on Split-Flow CascadeThis paper describes a traffic sign detection (TSD) framework that is capable of rapidly detecting multiclass traffic signs in high-resolution images while achieving a high detection rate. There are three key contributions. The first is the introduction of two features called multi-block normalization LBP (MN-LBP) and tilted multi-block normalization LBP (TMN-LBP), which are able to express multiclass traffic signs effectively. The second is a tree structure called Split-Flow Cascade, which utilizes common features of multiclass traffic signs to construct a coarse-to-fine TSD detector. The third contribution is the Common-Finder AdaBoost algorithm (CF.AdaBoost), which is designed to find common features of different training sets to develop an efficient Split-Flow Cascade tree (SFC-tree) for multiclass traffic sign detection. Through experiments with an evaluation dataset of high-resolution images, we show that the proposed framework is able to detect multiclass traffic signs with a high detection accuracy in real-time and that it outperforms the state-of-art approaches at detecting a large number of different types of traffic signs rapidly without using any color information.(3) A coarse-to-fine traffic sign classification methodThe main difficulties of traffic sign recognition are:a wide variety of signs, some signs are very similar, occlusion and pollution, light reflection and camera angle change. Based on the detection method, a coarse-to-fine recognition method is proposed, including the coarse classification method based on SVM and fine classification method based on Extended-SRC (Extended Sparse Representation Classification). In the coarse classification process, the first step is verification and the nest step is classification, which can significantly reduce the calculating time in coarse classification. The traffic sign classification method based on the ESRC is designed to classify traffic signs with small size dictionary. Instead of solving the sparse representation problem using an over-complete dictionary, the classification method based on the ESRC utilizes a content dictionary and an occlusion dictionary to sparsely represent traffic signs, which can largely reduce the dictionary size in the occlusion-robust dictionaries and achieve high accuracy. The experiments demonstrate the advantage of the proposed recognition approach and our TSR framework can rapidly detect and recognize multiclass traffic signs with high accuracy.(4) Occlusion-robust traffic sign detection and recognitionThe high variability of sign appearance with partial occlusions in uncontrolled environments has made the detection and recognition of traffic signs a challenging problem in computer vision. In this paper, an occlusion-robust traffic sign recognition framework is proposed. To achieve occlusion-robust detection, a color cubic feature called Color Cubic Local Binary Pattern (CC-LBP) is proposed to construct a coarse-to-fine cascaded detector. The CC-LBP utilizes color information and a self-adaptive threshold to express multiclass traffic signs, which can effectively remove non-object subwindows in the cascade-based detection. After detection, we design recognition method based on Extended-SRC to recognize traffic signs with occlusions. The verification experiments show that the proposed occlusion-robust TSR system can detect and recognize multiclass partial occluded traffic signs with high accuracy in real-time.(5) Multiview traffic sign detection and recognitionThe multiview appearance of road sign in uncontrolled environments has made the detection and recognition of road signs a challenging problem in computer vision. In this paper, we design a TSR system to detect and recognize multiview road signs. This method is based on several algorithms including the classical cascaded detector, the self-adaptive weighted Gaussian color model (SW-Gaussian model), and a shape context matching method. The classical cascaded detector is used to detect frontal road signs in video sequences and get parameters for the SW-Gaussian model. The proposed SW-Gaussian color model combines the 2D Gaussian model and the normalized red channel together, which can largely enhance the contrast between the red signs and background. The proposed shape context matching method can match shapes with big noise, which is utilized to detect road signs in different directions. After detection, the signs are resized into the same sizes using affine transformation. Then we use HOG+SVM recognition method to recognize traffic signs after affine transformation. The experimental results show that the proposed multiview TSR system can reach high accuracy in detecting and recognizing signs with different directions.