The Research Of Vehicle Logo Recognition Based On Gray Color Space

Vehicle Logo Recognition(VLR) is an important part of the Intelligent Transportation System and it acts as an important role in traffic management.The paper described license plate location algorithms and vehicle logo recognition algorithms.License plate location is the prior knowledge of VLR,so LPL algorithms were introduced firstly.For the current time consuming excessive location algorithm circumstances,a new algorithm based on statistical characteristics was proposed.The algorithm quantified the dense characters feature in the area of license plate as the jumping feature.We used dynamic programming algorithm in the searching process, effectively reducing the processing time by recording the intermediate results.In order to reduce the scope of the search,on the base of statistical characteristics,the localization algorithm based on banner searching was proposed.Using image edge detection to substantially distinguish the foreground area and background area.The foreground area would be significantly reduced.Then calculating the jumping feature values in the horizontal direction,so the foreground area would be further divided into separated banners.Merging and screening the banners,the searching area would be littler.Using the height of the banners as the height of the rectangles and the height multiplied by the ratio of standard width and height as the rectangles’width to search the banners.So the size of the rectangle could be determined dynamically. The method ensured the positioning accuracy and limited the time within15ms.For vehicle logo location,using the license plate tilt correction method(rotating projection method,narrow hole perspective model) to correct the car at first,and then get the corse vehicle logo location on the base of license plate location.Using the difference of projection between the horizontal edge image and the vertical edge image to get the logo background texture direction,then using marginalizing operation and morphological operation to suppress the background texture.Finally,the logo location was got by the analysis of histogram.For vehicle logo recognition,using the invariant of Hu moments and Tchebichef moments,the standard seven Hu moments and third-order Tchebichef moments was regarded as feature vectors,using KNN neighbor method as learning algorithm and the value of K would be changed in the experiments.Using histogram equalization,MSR and mixed method to enhanced the gray logo image.The experiments showed that when K equaled13,using mixed enhancement and Tchebichef moments as the feature vector,the recognition rate would be the highest and could reach97%.