Research On Traffic Sign Detection And Recognition Technology Based On Stable Feature

In the traffic environment, traffic signs are important facilities conductingtraffic management and ensure the traffic environment smooth and driving safety.Automatic detection and recognition of traffic signs is indispensable for intelligentdriving assistance systems and unmanned vehicles to understand the trafficenvironment. Therefore, the researches on automatic detection and recognition oftraffic signs have important research values and practical demanding. Lightingconditions and partial occlusion are two important issues in traffic sign detection andrecognition. It is a challenging research topic to detect and recognize traffic signsunder variable lighting conditions or partial occlusion efficiently. The main work ofthis paper includes:As the sensitivity of sequential scheme-based methods to lighting variation andpartial occlusion, a novel method is presented to detect traffic signs based on parallelscheme and contour geometric characteristic analysis. In this method, colorsegmentation and shape-based traffic sign detection are carried out on the inputimage in parallel, then the resulting regions from these two processes are grouped bythe OR operation to determine candidate regions. Finally, traffic signs can bedetected by verifying candidates using priori knowledge (e.g. region area and aspectratio). Experiments were carried out based on a public dataset including traffic signimages under diverse situations. The results show that the proposed method has anencouraging performance under variable lighting and partial occlusion.To improve the robustness of traffic signs recognition, a new method is proposedto recognize traffic signs based on SURFs (Speeded Up Robust Features) with highstability. A set of SURFs, which is called bag of SURFs, is constructed using SURFsfrom multiple video frames that contain a traffic sign, and is then matched with a setof SURFs from a standard traffic sign image. The recognition result is determined bythe highest score of a weighted scoring scheme. Experiments were carried out basedon a public dataset under three given conditions, and comparative analysis with tworecent methods was also performed. The results demonstrate that our method is anefficient traffic sign recognition method and yields a significant superiority undervariable lighting or partial condition.