Study On Recognition Method Of AGV Special Mark Based On Statistic Features

Due to the large information storage and highly intelligent level, computervision method has been applied to vehicle ideal navigation system widely. Afterhaving experimented and manufactured some model cars, my intelligent vehiclegroup has finally designed and successfully produced general AGV and assembledAGV which based on computer vision.During the process of navigation, vision navigated AGV can collect moreinformation and the information contained in its images is various, too, such asnumerals, characters, and special marks, which we often meet in our daily life andfactories, and so on. The abundant image information not only shows the highlyintelligent level of the vision navigated AGV, but also its strong flexibility.AGV can be applied in various situations, for example, to acceleration,deceleration, stop, turning in direct angles and recognizing T-cross, which are thefoundations to realize AGV recognition in many sub-branches. Therefore, theresearch on recognizing special marks is necessary. It's also one of the symbols thatreflect the superiority of AGV.The title of this dissertation is study on recognition method of AGV Specialmark based on statistic features. It contains five main parts: 1. an improvedthreshold segmentation method 2. Feature extraction of special marks 3. Templatematching in recognizing special marks 4. Recognition of right angle turning marksand T-cross turning marks 5. Arithmetic verification methodology and resultanalysis.The requirements of AGV in recognizing images are test and check on time,recognizing on line, and resolutions to practical questions including hardwaredesign and arithmetic design. The recognition methods of special marks whichAGV adopted before is based on line scan, which achieves a better effect in settledillumination, clean marks, and lower noise, but its reliability is weak in othercircumstances. Therefore it's harder to meet the requirements of AGV----the highreliability and high accuracy. In order to improve the reliability, I make a furtherresearch in recognizing special marks. Nowadays there are many threshold segment methods, but Otsu thresholdsegment is more classical. According to its theory, combine histogram and Otsumethods together, get an improved threshold segment methods. An obvious featurecompared with the Otsu method is that its celerity. Meanwhile the experimentsprove that this method can give accurate segment to AGV images. The information of some image features including contour features, stripfeatures, statistic features and the extraction and recognition of some fixed specificfeatures, is an important subject in computer vision and image process research.Experiments show that the image region edge is always correspondent with theedge of the target image, and the visual system of human beings often identifies theedge of targets in recognizing the image. With the requirements and conditions of AGV vision navigation as a reference,this dissertation introduces a simple way to get the boundary length which based onedge tracking arithmetic methodology. After getting the result of edge tracking, calculate the value of boundary momentby using edge pixel, then the feature value of the image is available. Furthermore,after get the information of a single pixel boundary of the target in interested region;I discover that as for Acceleration marks and deceleration marks, the hypotenuse ofthem and the arc of stop marks are features in pattern recognition. So we canrecognize marks according to the features of their shapes. During the process of recognizing special marks, feature extraction is also animportant content for some specific patterns recognition field. After calculatingboundary moment, observing and testing the value of boundary moment amongnine features, the differences among some data vectors in any category (totally 3categories) of the nine dimensions are not obvious. It means that in the ninedimensional data vectors, there is extra useless information, so we should compressthe dimensional numbers, that is, features extraction. The recognition methodology to different categories chosen in this dissertationis template match. Considering the reality of AGV in practice, I choose...