| In order to realize the sheep body measurement without contact, in this paper, thetechnology of measuring the sheep body size with machine vision and image processingwas studied. Through a lot of understanding of machine vision in the measurement at homeand abroad and the commonly used research methods, this paper did some research on thefollowing aspects about analysis and experiment validation: image preprocessing, contourextraction, the identification of measuring points and body size calculation.In image preprocessing part, removing noise and image shearing are studied.First,using median filter to remove the interference during collecting the image.Then,thecoordinates of four points were gotten by demarcating four vertices of backgroundplate.Based on the previous step,the image could be cut and lots of interferenceinformation could be eliminated effectively.In contour extraction part,Sobel-Hough lines extraction, envelope extraction andcurve fitting were mainly studied.In this paper,the designation strip of black and white wassticked on the background.In this way,through analyzing the designation strip,the back ofthe sheep body contour points were gotten.First,through Sobel based on the first-orderdifferential operator the edge was detected.Because the designation strips were in thevertical direction,only the convolution operator in the horizontal direction was selected.Theadaptive threshold was used to split the edge information.With Hough the line could beextracted from edge information.After image thinning,through column scanning the backof the sheep body contour points were gotten.These points were connected as a curv and it was smooth by Bessel fitting.In measuring point identification part,after the analysis of a large number of sheepsamples,the characteristic information of measuring points were gained.Among them, thehip point has the largest curvature.The point could be selected in D-P and Helen chiu-shaoformula.After the analysis of a large number of sheep samples,In this paper,a method wasproposed that the key frame concluding the ideal posture of sheep could be recognizedautomatically.Through the length of the line called L connecting the head top and the hippoint and the angle called α,whether the posture ideal or not could be ensure. Imagebackground plate calibration vertices was regarded as sheep hoof point.In body measurement part,through analizing the relationship among the image coordinate system, the world coordinate system and the physical, the spatial resolution wasdetermined.After correting error,body measurement was calculated.Through the experiment of median filtering, image shearing, Sobel Hough linesextracted, contour extraction, the extraction and body measurements of measuring pointcalculation for validation,some conclusions below could be gotten. Under the condition ofnatural light, after the image sheared, most of the sheep body key information could beacquired.Meanwhile,most of the interference was removed. The influences of light wereremoved in the way of background identification and complete back contour was obtained.According to the L and alpha two parameters,the ideal posture was choosed accurately. Hipand shoulder point could be obtained accurately.The error of height and length of thesheep body was less than5%. |
