Study On Mature Fruit Location And Obstacle Detection For Citrus Picking Robots

With the rapid development of agricultural production,the cost of agriculture labor force will become more and more costly.In recent years,the agricultural application of robot technique have already become popular issue,because of the shortage of the agriculture labor force both in developed countries and developing countries.Different from industrial robot,which works in particular environment,the agriculture robot mainly works in the natural environment,and the agriculture robot has to face more complicated and uncertain circumstance,thus there are more problems to be resolved.As a part of research on citrus picking robots,this research used binocular stereo vision to researching on recognition and location mature citrus,obstacle(branches) detection under natural environment.The main contents and methods are as follows:1.Mature fruit locationThe main steps to complete that are image recognition and stereo matching.The purpose of image recognition is recognizing mature citrus region from image to make preparation for location.This research used iterate on 2R-G-B chromatism component in RGB color system to finding threshold automatically to segment original image. Then the segmented images were converted to two-value images and eliminated noise by morphological operation.Region labeling was done and eliminated region which was small or had great differences from circle by defining threshold which based on area and length to width ratio of the smallest circumscribed rectangles of each region. Then region filling and contour extraction operation were been done and used improved Circular Hough Transformation(CHT) to found out circle centre and radius of each region's approaching round.Then used feature-based match and added extra characteristics such as barycenter of orange region in the image,sizes of each region's approaching round to get correct match results of each image pair.Experimental results show that the matching accuracy can reach over 80%.Finally,after calibrated the camera interior and exterior parameters,the 3-d space coordinate of each orange was been obtained and used the laser range finder to verification and comparison. Results show that the average error ratio is below 1%when the measuring distance is no more than 1.5m.2.Obstacle detectionTo ensure executable of algorithms,the steps of obstacle detection are the same as that of mature fruit location.Used iterate on 2R-G-B and 2G-R-B chromatism component,combined with gray threshold method to segment image quickly and effectively.Got the branch regions by image binaryzation,morphological processing, region labeling and filling.Extracted skeleton of obstacle by thinning and did some processes so as to pruning the skeleton and recovering the occluded skeleton.Then obtained the feature points such as endpoints and branch points of the skeleton, recorded their connecting relationship.Finally the 3D information of obstacle was restored by stereo matching on feature points.Experimental results show that the identification accuracy of obstacle can reach 67.3%,the identification error ratio was increased when the actual distance of obstacle is more than 1.5m.Through the research,some achievements have been made.Such as mature fruit recognition,match and location.This research also provides a method for fruit harvest robot to detect obstacle.The research results of this research have reference value for the study on visual recognition in the field of harvest robot in our country.They also provide a basis for further study and have important economic significance to enhance international competitive power of our country's agricultural.