Detection of Street-Level Traffic Panels Based on Cascaded Color Segmentation

Abstract

Traffic panel detection and recognition is an important part of Intelligent Transportation Systems (ITS). However, owing to a variety of complex natural scenes, traffic panel detection and recognition still remains a challenge in computer vision. We propose a robust approach for traffic panel detection from street-level images. Considering the visual appearance of traffic panels, we first extract candidate regions of the traffic panels after applying the cascaded color segmentation method we propose. The candidate regions are selected by analyzing a bounding box of contours in the segmentation binary image. Second, the features of the candidate regions are extracted, and the descriptors are computed. The traffic panel images are represented as a “bag of visual words” by clustering the centers of the descriptors. Third, the traffic panel images are detected using support vector machine classifiers. Experimental results on images from Tencent Street View indicate the effectiveness of our proposed method.