Research On Three-dimensional Reconstruction Technology Of Rice Seed Based On Computer Vision

With the rapid development of agricultural information industry in China,applied research increased in more and more computer vision technology fields,the trend of computer image processing technology which applies in agriculture development has improved significantly.Rice seed surface morphology is an important aspect of seed purity identification and recognition.Considering that artificial recognition and identification methods have a number of faults,including low efficiency,high labor costs,and poor accuracy,scientifically selecting quality rice seeds by using computer vision methods is important.Different models and methods have been established in the field of crop seed identification.Studies on rice seed speciation analysis methods indicate that current detection methods in computer vision mainly analyze 2D information and that the use of 3D models is lacking.This paper proposes a 3D rice seed reconstruction system which can be used to measure the morphology of rice seed,with more accurate shape measure parameter values and more comprehensive appearance character and defect expression.In this paper,a new crop seed reconstruction system which supports fast and accurate recognition was designed to build a 3D surface morphology.The Depth-From-Focus(DFF)method was applied in the analysis of crop surface morphology.The Fast Point Feature Histograms(FPFH)algorithm was applied to define characteristics of point cloud.The RANdom SAmple Consensus(RANSAC)algorithm and Iterative Closest Point(ICP)algorithm were used to support the research on point cloud registration.In order to calculate characteristic value,tetrahedrons method and triangular patch algorithm were applied to calculate volume and area of rice seed respectively.New method was designed for fast registration algorithm with characteristics of block fusion.Three-dimensional reconstruction with feature value measurement system was realized.Precision of the measurement reachedμm.Variety recognition system of rice seed was design with database which supports rice detecting by using neural network modeLResearch contents and results were summarized as below:1.A 3D reconstruction and feature measurement experimental platform was developed as the experiment platform.The design criterion of the platform is stabilization,high precision,fast response and easy to control.The platform was composed of hardware system and software system.Hardware system consisted of Optical platform,vertical lift module,camera module,light source control module and data comnunication bus.Software system consisted of image preprocessing module,image acquisition module,three-dimensional point cloud producing modules,feature extraction module,point cloud registration module and feature value measurement and varieties identification module.Experimental platform performed steadily and reliably data communication.2.Verify the best combination plan of camera and lens applying to the work of image acquisition.Comparison By analysis of the three kinds of visual equipment,the most suitable image acquisition device plan was given.Industrial camera and tele-centric lens has the advantage of low distortion and high precision.through the analysis of the European space,with the camera imaging surface pixel coordinates and the spatial point world coordinate transformation,the relationship between monocular camera calibration method was given.Through the calibration experiment,lens’ inside and outside parameters were calculated by using internal and external parameters of the projection transformation matrix.The experimental result showed that the average absolute error of measurement was 0.003 mm,between 0.001～.0.010 mm,the average relative error measurement was 0.05%,between 0.02%～0.15%.3.A method for three-dimensional point cloud reconstruction with DFF algorithm was proposed.The depth-from-focus(DFF)method was applied in the analysis of crop surface morphology.Image sequences were acquired by using a specific vision device in which the distance between the camera lens and the rice seed differed.High-pass filtering was used to extract pixels and analyze strength value changes in the frequency domain.The second-order differential was employed to strengthen the value in.the frequency domain by using the improved Laplacian operator.Threshold statistical analysis was conducted in pixel windows,by which each pixel generated a value which shows the focusing condition.The focusing measure of the image sequence effectively determined the estimated depth value of a pixel,and a focusing pixel stack could be defined based on these values.Using the characteristics of the Gaussian distribution of the focal depth estimation value,Gaussian interpolation was calculated to obtain a more precise surface morphology depth value.As a result,a depth image collected based on the estimated depth value of the pixel was developed.Finally,through depth image smoothing and edge pixel processing,a 3D point cloud could be produced.Thus,a rice seed reconstruction system which can be used in rice seed identification and recognition was designed.The result showed,when optimized point cloud with statistics window of 5×5 and threshold value of 7,the average absolute error of measurement was+5μm,the average relative error measurement was 0.07%.4.A method of characteristics block fusion for three-dimensional point cloud quickly registration was proposed.In the traditional algorithm of point cloud registration,the initial registration algorithm are combined with precise registration algorithm,with which accurate registration of point cloud could be realized.Main faults of traditional three-dimensional point cloud registration method are high complexity of algorithm,low efficiency and easy to produce a large number of redundant data.In this paper,advantages and defects of traditional method are summarized,the quickly registration algorithm of block point cloud which based on random sampling algorithm(RANSAC)was proposed.This algorithm only need to extract the cloud data in two parts,while the point cloud is extracted from two sequence of images in 180 angles.According to the characteristics of data blocks,random sampling matching values could be found.The initial relationship which supports the initial registration of rice seed was built by using the matching value estimate.Finally,with the initial registration of point clouds which iteration 5 times by the ICP registration algorithm,a satisfied three-dimensional point cloud model could be realized in less than 60 seconds time and the precision grade was μm.5.Three-dimensional reconstruction and measurement system was designed and realized.Ten size of key characteristic value of shape were obtained.Rice seed shape could be reshaped by two-dimensional and three-dimensional measurement value.In order to calculate characteristic value,tetrahedrons method and triangular patch algorithm were applied to calculate volume and area of rice seed respectively.The characteristics of the corresponding morphological analysis is used to variety identify of rice seed.Neural network identification analysis database was established.Three-dimensional reconstruction system can effectively overcome several deficiencies of traditional seed speciation analysis methods.Through further calculate,the surface morphology characteristics of seed were shown.The new 3D surface morphology could serve as an important reference for researchers.Finally,BP neural network model was structured to support variety identification,Suitable neural network algorithm was selected for five different sorts of rice seed,and the final identification rate was between 80%and 100%.