Recognition And Localization For Apples Under Natural Environment Based On Binocular Vision

In order to solve the problem of labor shortages and rising labor costs,it is necessary to research and develop fruit and vegetable picking robot,and the key point in the study of harvesting robot under natural environment is the identification and localization of fruit and vegetable.In this research apples are regarded as samples,image segmentation methods for apples,identification methods for clustered and obscured apples,and location methods for apples were studied by using stereo vision technology.Some problems in recognition and location were solved in this paper.The main research contents and conclusions are listed as follows:Through summarizing the regularity of the chromatic aberration method used in apple image segmentation,a method based on iterative method to determine the dividing criterion was proposed in this paper,and the scope of the iteration was determined at the same time,and then by making iterative standards,the segmentation criterion was determined.100 groups of segmentation experiments under different light intensity proved that the segmentation criterion which determined by iterating different light intensity apple images can segment the apple out from different light intensity image.A method according the angle between two of the three equidistant points in the apple contour's pixels for extracting the apple's normal contours was obtained by analyzing the distribution of apple contour's pixels.And then clustered or obscured apples were recognized by fitting apple's normal contours by the least square method.Through experiments this method was compared with the method based on curvature to extract the apple's normal contour,and the results show the method based on the angle between two of the three equidistant apple contour's pixels has better recognition effect,the the correct recognition rate reached 89.5%.The binocular cameras were respectively calibrated by functions in OpenCV library computer vision and calibration APP in Matlab2015 b.Calibration precision was improved by adjusting the parameters in OpenCV calibration functions.The two methods were compared by calibration experiment and the results show that by setting the samedistortion model the calibration precision based on OpenCV is slightly lower than based on Matlab2015 b.Multiple apples exiting in left and right images were matched by proposing matching criterion,and then the location of apples was respectively realized by least square and parallax method.The two location methods were compared by 20 groups of location experiments,and the results show that the locating method based on least square method is with higher positioning accuracy and real-time performance.Finally the real-time positioning of apples was implemented by coding program based on MFC dialog.