Study On Sheep’s Body Size Parameters Measurement Extraction And Three Dimension Reconstruction Based On Binocular Stereo Vision

Sheep body size parameters and weight are the key indicators to measure the growth status and production performance of sheep.The traditional measure way to sheep’s weight and body size is to use stick and sacle.The traditional body size and weight of sheep are directly measured by using tape,stick and scale.This way not only is a large workload,but also generates stress response,leading to reduce the welfare level of sheep.For the above problems,live sheep and specimen sheep are regarded as the research object.The dissertation mainly studies sheep body size parameters extraction algorithm and three-dimensional reconstruction based on binocular stereo vision.The key research contents,research methods and research conclusion are listed in the following.(1)Study on camera calibration.The captured calibration board images were implemented image preprocessing,target extraction,target property calculation and denoising processing,so as to accurately obtain the features of calibration circles.The intrinsic and extrinsic parameters of camera were calculated by adopting Zhang calibration method.The calibration results compared with the commercial binocular measurement platform results,and were analyzed by using image coordinate residuals computed by three dimension space reprojection.The results show that the calibration results are better.(2)Study on sheep body segmentation.Under the real complex breeding environment,the improved Graph Cut sheep body recognition algorithm was proposed,which combined the multi-scale watershed and fuzzy C-means algorithm to enhance the efficiency and accuracy of segmentation.From three aspects,including segmentation effect,interaction time and segmentation accuracy,the proposed Graph cuts algorithm was verified,whose average interaction time is 6.47 s and measurement value F is 0.97.For the overexposed images,the multi-scale Retinex algorithm with color restoration was used to enhance image.By the tests for multi-group images,the proposed algorithm can precisely detect sheep body,which is more stable.(3)Study on the estimation model of sheep body weight and body size measurement points.The independent variables for estimation of sheep weight have the correlation and collinearity,which include body length,height at withers,and rump height.On the basis of Pearson correlation analysis,we employed the stepwise regression and partial least square regression method to build sheep weight estimation model.The goodness of fit R2 of the former was 0.9321.However,the model didn’t contain the body length parameter that have strong correlation with weight.According to correlation analysis,partial least square regression method can more describe and explain the weight prediction model.The prediction model based on body length,height at withers and rump height was constructed,and the average relative error was 3.17%.Then,on the premise of image segmentation,the body contour line was extracted and contour area was divided into four parts.The sheep body measurement points based on envelop analysis were identified,to realize eight measurement points extraction.These points were the points of body length,height at withers,rump height and chest deep and so on.The measured values based on measured points of two dimension image compared with the real values,the average error is less than 2 cm,and the relative error is 1.95%.(4)Study on the three dimension sheep body reconstruction based on binocular stereo vision.Aim at the three dimension body size parameters,such as the chest and circumference of cannon bone,the two dimension image cannot measure.Using the binocular stereo vision technology,we extracted the SIFT features of sheep image under complex background.For RANSAC algorithm has the strong performance of estimation fundermatal matrix,the improved RANSAC epipolar constraint feature match algorithm was proposed.Due to collecting the sparse three dimension feature points,the point cloud interpolation algorithm was applied to obtain the dense point cloud.The NURBS surface fitting can fulfill the three dimension sheep model,so as to extract the body size parameters.The relative error is 3.47%,which is slightly larger.(5)Study on the method of active three dimension sheep body reconstruction compared with the binocular stereo vision.Firstly,sheep point cloud data were denoised by the improved algorithm of k-nearest neighbor.Then,octree coding was employed to reduce data,which can get evenly distribution of point cloud data and retain sheep features.Finaly,the three dimensional surface model of sheep body was fast and accurately reconstructed using Crust algorithm,and the parameters were extracted.Compared with the real values of body size parameters,the average relative error of the measured values is 1.33%.The detected values by three different approaches made the comparation.The body size detection accuracy of the laser scanning three dimension reconstruction is higher.However,this way are limited and the laser beam is harmful to sheep body.The accuracy of two dimension image is slightly lower.But it cannot detect body width.The error based on the binocular stereo vision reconstruction is a little larger.However,the error is within the range of error permitting,This method can meet the sheep body size parameters measurement.Thereforth,sheep body three dimension model based on binocular stereo vision reconstruction can be used as the first choice method for measuring body parameters.