The Intelligent Vehicle Logo Recognition System In Road Bayonet Surveilance

In the fast-paced city life, the popularity of cars has brought us greatconveniences. However with the increasing number of vehicles and theexpanding scopes of monitoring, manual video surveillance could not meetthe needs of modern social development. The intelligent management ofvehicles has become an urgent issue to be solved. As an important mark ofvehicle, the vehicle logo recognition has been a hot topic in the roadsurveillance.With the widely application of HD monitoring technology in roadbayonet, vehicle logo detection and recognition in road bayonetsurveillance system have great values. This paper will introduce ourresearch on the vehicle logo detection and vehicle logo recognition in theroad bayonet surveillance.Based on the analysis of the current vehicle logo detection solutions,this paper proposed a new vehicle logo detection system. We adopted acoarse-to-fine detection method. In the phase of coarse detection, to assistlogo detection, the license plate is first located based on the information ofplate edge detection, texture projection and etc. In the phase of finedetection, we proposed a vehicle logo detection method based on thecombination of heuristic discover and texture recognition. This algorithmcan be applied to the cars with different background textures andnon-symmetrical structures.Considering the characteristics of the vehicle logos and the limitationsof existing vehicle logo recognition methods, this paper proposed a newvehicle recognition algorithm based on the Bag-of-Words and SpatialPyramid. Firstly, extract the Dense-SIFT features of logo images. Then build a visual vocabulary by K-MEANS and quantize the descriptors ofsample images into visual words. Thirdly, introduce the spatial informationof the visual words by Spatial Pyramid model and represent the logo imageas a spatial visual words histogram. Finally, Support Vector Machine(SVM) is adopted to classify the vehicle logos.The proposed vehicle logo detection and recognition system isevaluated on large amounts of the road bayonet surveillance images andother datasets. Experiment results verify that the proposed one with a highaccuracy and a fast speed.