Research On Processing Method Of Orange Trunk Recognition

The research on auto navigation by vision is one of the focuses of computer vision in recent years.This paper is a basic part of Orange Harvesting Robot's auto navigation work.The existence and position information of the orange trunk indicated those of the Orange-Tree in this paper.The method of orange trunk image segmentation and feature extraction is to segment an individual orange trunk from its surrounding in the photography image of the orchard,and mark the feature points.This paper,combining the features of orange trunk image self and known knowledge,proposes two new methods on the basis of the known arithmetic that is suitable to segment the orange trunk out of image.At last,automatic segmentation of orange trunk images were realized.The following are the main research works and conclusions.1.Try to segment the orange trunk image by three methods of known image segmentation algorithms,analyze and contrast their results.To draw a conclusion,it is unsuitable to orange trunk image segmentation by simple using the known algorithms.2.A new arithmetic based on color image segmentation to detect orange trunk was given.After the orange trunk image analysis,the ratio of G lightness value to R lightness value of pixels in the orange trunk region is about 0.9.By initializing the color feature vector for color segmentation,the orange trunk was extracted by main processing steps of linear filtering and shape features segmentation.The discontinuous region was repaired by mathematical morphology,which completely fills the holes on orange trunk and have a good effect on connection the region of fracture.Recognition success rate is up to 75%.3.This paper puts forward a Peeling Algorithms that is suitable to segment the orange trunk out of image.An orange trunk image can be divided into three parts: orange trunk,green region and background.The peeling algorithm idea is to hold on the orange trunk region by removing green region and background,the orange trunk was accurately segmented from an image.The algorithms have been successfully applied in removing orange crown,weeds,sky background and light spots by pixel point lightness value operation,but hardly change the information of orange trunk region.On binary image,this method removed soil background by using characters between trunk and soil background and truck.The experiment showed that recognition success rate is up to 92.5%,recognition rate of orange trunk,on the average,86.2%,and the average processing time 1.535s.After putting the orange trunk back to the original image,the location error of orange trunk medial axis reached 0.802 pixel on the average.This research gets the new methods of orange trunk recognition,which can provide the research base for Chinese Orange Harvesting Robot.It provides a new idea for pushing forward the automatic scheduling for fruit robots.