Research on Recognition of Traffic Signs from Natural Scenes
|School||Agricultural University of Hebei|
|Keywords||recognization of traffic signs signs positionsing image segmentation wavelet neuralnetwork|
With the rapid economical development, The automobiles have entered innumberable families.On one hand, cars make our travelling more convenient; on the other hand, the frequent trafficaccidents caused by cars harms the victims’ family greatly. Therefore, studying the inteRgenttransportation technology can facilitate drivers’ travel and reduce traffic accidents.As an important constituent of intelligent transportation system, the traffic sign recognitionSystem functions as follows: first, the traffic sign recognition system shoots natural scene images bythe camera installed in the front of the car; then, the traffic sign recognition System transfers theimage to image processing module to identity traffic signs; finally, the traffic sign recognition Systeminforms the driver of the recognition results through voice or display screen, which can make thedriver drive safely. Over ten years, the study of traffic sign recognition has made great progress andhas made some achievements, However, compared with targets recognition system in unnaturalscenes, targets recognition system in natural scenes is more challenging due to the existence of thecomplex backgrounds, light and other various factors. In order to improve traffic signboard’sdetectable velocity, Recognizable Speed and accuracy in natural scenes, in this thesis, the author takesComputer Image Processing Technologies as the main technological means, applies HSV-color spacemodel, image preprocessing, Affine transformation correction, shape placing, Feature ExtractionBased on Moment Invariants and wavelet neural networks Comprehensively to study automobilerecognization of traffic signs, which have gained better results. The main contents of this paperare as follows:1.Traffic Sign location based on color and shape in HSV spaceEvery Traffic sign has its own specific color and shape, in this paper, the author classifiesdifferent traffic signs according to the traffic sign’s color and shape into directional signs, prohibitionsigns, warning signs,"Stop and give way" sign and "reduce speed and give way" sign. Forexample, the round oval red speed limit signboard’s outer border is white background with black text.the triangular warning sign is yellow background with black text.2.Traffic signs’ image preprocessing and Affine transformation correctionIn this paper, the author carries on Color pretreatment on the collected images in natural scenesusing white balance and color enhancement algorithms and corrects the anamorphic traffic signs innatural scenes with affine transformtion technology.3.Texture feature extractionThe traffic signs of the same class are distinguished by Texture feature extraction, in this paper,the author takes Hu invariant moment and Zernike invariant moment as two feature vectors to identifytraffic signs.4.Traffic sign recognitionTraffic positioning distinguishes the types of traffic signs, it is acceptable just to identity trafficsigns of the same class when you are identifying the traffic signs. In this article, the author trainswavelet neural network and BP neural network to identify individuality of every category based onthe extracted Hu invariant moments’ and Zernike moments’ invariants, whose recognition rate canreach above90%. The experiment has verified the validity of the method.