Research And Performance Analysis For Algorithm Of Traffic Sign Detection

The real-time detection and recognition for traffic sign is an important component of Intelligent Traffic System (ITS). And as more and more organizations and companies competing research and development of unmanned vehicles, this technology received more and more attention and study.In this paper, the establishment of database for traffic signs and traffic signs in real-time detection are studied. Including red, blue and yellow types, the TSR system can recognizes24species of traffic signs. Collection of experimental data and algorithms on the test are based on the experimental for unmanned vehicle. And our Traffic Sign Detection and Recognition System (TSR) can adapt to the changes of weather, light and perspective. The specific research is as follow:1. At home, study on TSR at this stage is lack of publicly available traffic sign image database. With analyzing the characteristic of traffic signs’image obtaining, this paper established a traffic sign image database which includes detection image test set, recognition training and testing sample set.2. Currently, there are so little methods to evaluate the performance of different algorithms of Traffic Sign Detection. Therefore, an approach to analyze the effect of detection algorithms which is based on TSR system, are proposed. The mechanism of this method is same to that of black-box testing. By fixing all models constantly except the model of detection algorithms, use testing image set to examine different ways of detecting, and evaluate the statistic data including output rate of detection, recalling rate, output rate of recognition, undetected rate, unrecognized rate and computing speed of system.3. Because the failure of detection usually happens in bad conditions, for instance, dim light, traffic signs are leaned and so on, color image enhancement based on MSR(Multi Scale Retinex) is designed to detect the targets, and LUT(Look-up Table) technology is utilized to accelerate the algorithm. In this approach, two kind of exploring ways are applied. Firstly, the first time explore is done, then segment the foreground by OR computation of OST/SVT, and LUT technology is used. Secondly, project the first exploring results into source images; color enhancement is realized by MSR, and in the end use the shape information to obtain the detection result.