The Container Character Recognition Based On Machine Vision

With the development of the transportation, container, as the mainway of freight transportation, is playing a more and more important role.Unfortunately, although the modern management mode has been adoptedby many container terminals, the identification of container charactersand the container check still need to be read and recorded by manualways, which greatly influence the speed of container transportation. Theautomatic identification system of container characters is one of the keyto solve the problem.The automatic identification system of container characters base onmachine vision and artificial intelligence has been researched by manyscholars, but few of them can combine the system with the actualapplication. The main reason is the container images in actualapplications are very complex and many problems need to be solved. Inpractice, the recognition system of container characters is complex, whichcan be mainly described as follows:1. The position where the containertruck park can hardly be fixed, so the view of the image can hardly befixed, which lead to the distortion of the container image.2. The light condition is different when in the night, in the morning or at dark, what’smore, the strong light often occurs in the white, which makes the imagecomplex.3. The rain or snow weather, and the polluted surface ofcontainers will interfere the system.4. The image resolution is relativelylow because of the filming equipment.This paper mainly focuses on the practical application to design thewhole recognition system of the container characters for the white image.This paper has researched the automatic recognition system of containercharacters, and mainly focuses on three main parts including containerbody detection, container character extraction and container characterrecognition.In the part of container body detection, we first correct the imagedistortion through the perspective transformation. Then, by the method ofRGB model to judge whether the container is color or white. Finally,detect the color container body through HSV color model and detect thewhite container body through the Scharr filter of texture, and obtain thefinal detection result.In the part of container number extraction, we preprocess the imagemainly by the machine vision methods, such as image graying, edgeextraction, mathematical morphology and binarization, then, get theregion of container number by the methods such as detecting andselecting enveloping rectangle, histogram projection, finally, segment the character and reprocess the segmentation according to the arrangementcharacteristics of container number to increase the robustness.In the part of container number identification, we mainly recognizethe container number base on sparse representation, getting the sparsecoefficient solution by the method of L1/2, then computing the residual ofthe origin image and the reconstructed image, finally classifying theimage to the category who has the smallest residual. We also comparethis method with the method of template matching to reflect the highrecognition rate and robustness. We also try to use the method torecognize the distorted container number in highlight.In this paper, we proceed the identification experiment, and thetested container image is collected in Shanghai Waigaoqiao customs. Theexperiment shows that the recognition system of container body base onmachine vision can effectively identify the container characters. Therecognition rate of single character is above99%, and the wholerecognition rate of the system is89%, so the system has very goodapplication prospect.