Study On A Seedlings Separation Method Based On 3D Vision Measurement Of Cotyledon Contours

Grafted seedlings can be improved ability to adapt to the environment, and improving quality, it is important for agricultural production.Matching relation between grafting on seedling root stocks have a certain requirements in order to improve the survival rate of grafted, but automatic and semi-automatic grafting machine did not do this..We need to measure the data of the size of seedlings and root stocks in order to achieve dish matches of automatic grafting machine for fruits and vegetables.It will be have a condition of occlusion by multiple strains cotyledons when detect the seedlings on graft. It will bring great disturbance to visual inspection and must be separated.Based method of binocular vision to achieve plug cotyledons profile measurement of three dimensional data and using cotyledon profile data at the height of the separation of the cotyledons of mutations blocking each other, laying the Foundation for the subsequent seedlings and root stock match. This research work mainly in the following areas:1、Construction of the binocular vision system.Building a portfolio of binocular vision system, and using a binocular vision system based on concentric circles detection online comprehensive adjustment method for system error control.By this method you can effectively adjust the optical axis of binocular system, the angles of camera and binocular camera focal length, making the system more reasonable and lay the Foundation for subsequent matching.2、System calibration based on concentric circles detection.Using a camera calibration based on concentric circle detection method which is based on the advantages and disadvantages of the analyze of the aberrations of the imaging model and calibration method.The method set the center point of concentric circles as the the characteristic points instead of the checkerboard corner in zhengyou Zhang method.By mean of the concentric circle makes feature point extraction more accurate than the checkerboard corner.The experience proves that the error of concentric circle calibration is reduced by 9.6% and the speed is up to 14.8% than the checkerboard calibration.At the same time analyse the epipolar geometry which proves that the epipolar constraint would make the feature matches from two dimensions to one dimension,which greatly improves the speed and accuracy of feature matching.Experiment shows that by the polar search,the error of potential and real points is about 2 pixels, we can detect the exact match point through the match of eight neighborhood points.3、Cotyledon outline study on measurement of three dimensional data.At first using the connected component labeling method for image denoising and enhancing the image by the method of color difference method and achieving the seedling cotyledon separation from background by means of the Binary conversion.Using the connected component labeling based on run-length to get the outlines of seedlings and separate it by the difference of connected components and down-sampling the contours sequences by Freeman chain code.At the same time,use gray scale vector method based on window (Gray Vector), simplify the SIFT descriptor (SSIFT), circular pattern SIFT descriptor (Circle SIFT) three matching method for binocular image contour matching.At last,calculate the 3D coordinates of contour series by the method of triangulation principle.Import the contour information to Matlab for seedling cotyledon of three-dimension reconstruction.Experiments show that the fastest SSIFT descriptor single point takes 0.79ms,Gray Vector followed by the descriptor, single point takes 0.98ms,Circle SIFT descriptor is the slowest, single point takes 4.17ms in the matching speed;Gray Vector descriptors is the most, followed by SSIFT descriptors, last is Circle SIFT descriptors in match points; Circle SIFT descriptor on the highest in matching accuracy.But considering the pair three dimensional reconstruction of blade profile,fewer points against the contour representation of three dimensional information, we use the Gray Vector descriptors to match.4、Study on cotyledon separation method.Using three dimensional data filtering removes the the isolated mutations points’effect which is created by contours matching offset caused by the contraction contours.Data filtering is mainly look for the point that two points around the it there is a big difference in height.But if only side there is a big difference between the potential is assumed continuous mutation point.Chain marker method was used to mark the points in a row, set relevant thresholds to merge adjacent chain to eliminate the error caused by a number of short chain on the effects of seedlings separation.Taking into account the same cotyledons large three dimensional jump between successive points will not exist so as to separate individual leaves in the seedling. Experiment on plant seedlings, two strains of occlusion-free seedling,two strains of occlusion of the seedlings for the detection or separation, while shielding cotyledons are rotated in 45° direction to verify the robustness of the algorithm.Experimental results show that from a data filter,contour link tag, chain merge method can effectively detect the seedling cotyledons and separation.