Research On Recognition And Extraction Of Road Nameplates Based On Vehicle Serial Images

Road brand information is one of the important components of urban geographic information.Accurate and high precision road brand information plays an important role in urban planning,traffic control and emergency response.In the "digital city",the upgrading of road brand is also an important part.The target recognition of sequence image is a frontier research direction in the field of computer vision.The recognition of target by sequence image is an edge science of computer vision technology,image processing technology and human intelligence technology,that is to analyze the image processing method of the target.Qualitative analysis and quantitative photogrammetry.In this subject,the target is detected and identified in the image sequence obtained by the digital camera carried by the vehicle mobile measurement system,and its behavior is understood and semantic described.The fast and accurate detection of road brand is an important problem to be solved in traffic signal recognition,and it is also an application research direction of multi disciplines,such as image processing,pattern recognition,machine vision and so on.However,with the influence of complex and changeable natural environment,the real-time requirement of the system,the high-speed motion of the vehicle and the processing speed of the hardware,there is still not a perfect and perfect detection scheme so far.That is to say,it is very difficult to detect brand information quickly and accurately.In this topic,based on the vehicle serial image,with the help of MATLAB,python-OpenCv,Anaconda3 and other software,this paper discusses the algorithm of road brand logo positioning.The principle of this algorithm is to locate the name of the name brand by combining the target mark rough location based on the color space and the accurate location based on the shape of the region.The task of target mark detection is to find out all the regions containing the target mark from the input image.The color based detection algorithm proposed in this paper aims at the characteristics of the road brand logo.By detecting the blue threshold in the image,the background color similarity and the existence of a variety of symbols are solved first.The shape based detection algorithm locates the rectangular sign by detecting the rectangle shape in the image,and it is based on the shape based detection algorithm.The template obtained before threshold processing looks for the highest matching area in the approximate sign area,that is to get the road brand.Experimental results show that the matching algorithm based on color and shape features has higher detection accuracy and satisfactory results.In addition,deep learning is used based on Tensor Flow framework,and SSD+MobileNet model is used to detect,identify and extract road nameplates.This method is more suitable for big data road nameplates.