| The SIFT is an algorithm based on scale space, invariance with changes with image size, rotation, or even affine transformation. Recent years, scholars have begun to use SIFT algorithm in the traffic signs detection and recognition. However, if directly applied to the traffic sign detection and recognition, the generated large number of unrelated feature points will reduce the efficiency of the algorithm, and cause the instability of the algorithm. Moreover, the classical SIFT algorithm descriptors are generated only by the grayscale image, ignoring the color information, resulting in the poor efficiency of image matching. SIFT algorithm based on color image, such as HSV-SIFT, OpponentSIFT, RGB-SIFT and the W-SIFT, first calculates the three-channel 128-dimensional feature vectors of the image, and then synthesize a 128* 3-dimensional characteristics vectors, finally achieve the goal of matching of color images. However, light changes cause low efficiency of these algorithms, and the feature vector dimensions are more than classical SIFT algorithm, resulting in the time consuming of the algrithm. Therefore,it’s difficult to use these algrithms in traffic sign detection and recognition in the actual scene.In order to reduce the interference of the invalid feature points, before using of the SIFT algorithm, the algorithm based on detection-recognization framework, uses color features to detect the ROI area of the traffic signs first;-in order to increase the invariance of the algorithm, we remove Non-one mapping matching points, and calculate the angle of the feature points and the center of the image, remove the angle difference too large matching points.Experiments show that with ROI calculation and delays screening can make the SIFT algorithm has a better efficiency to identify.traffic signsTraditional traffic sign detection algorithm based on color features will become invalid in the case of illumination. The paper presents a dectection algorithm fused with brightness feature and saturation feature of the traffic sign, combined with the dynamic brightness threshold and saturation threshold,also color information to detect the traffic signs. The paper proposed an improved color histogram that with brightness and saturation, ignoring saturation and brightness of smaller pixels, so as to improve the stability of the algorithm. Based on the above, a color histogram based on the improvement of the probabilistic model and the SIFT algorithm is proposed to achieve the matching of color images. The experiments show that the traffic sign detection method can accurately detect the illumination of traffic signs; compared with the traditional SIFT algorithms, the SIFT algorithm combined with the improved color histogram is more robust and has better recognition properties. |
