A Method For Obtaining Target Fruit Coordinates In Space Using On Kiwifruit Picking Robot

The acreage and production of China’s kiwi fruit rank first in the world. However, at present, the kiwi fruit is mainly harvested manually, which is highly labor-intensive. With the development of science, it becomes possible for kiwifruit picking robot harvesting instead of people. Fruit spatial coordinate acquisition not only is the prerequisite for kiwifruit picking robot but also a key technique for Mechanical-picking.The aim of machine vision system lies in the fruit image recognition from disturbances like branches and leaves to achieve precise position of fruit and finally realizing fruit picking. Compared to industrial robots, agricultural robot has higher requirements on fruit recognition and spatial coordinates’ acquisition due to the changing working environment and complex disturbances. Therefore, target fruit spatial coordinates acquisition method of kiwifruit picking robot is of great practical significance and scientific interest.This paper doing research around the fruit spatial coordinate acquisition with clusters Hayward kiwifruit in trellises cultivation pattern as object. The main research contents are as follows:(1)Kiwi fruit recognition. Target fruit recognition is the prerequisite for fruit picking robot. The effect and method of recognition determined the picking robot morphology and productivity in some extent. This paper did survey at Kiwi Experimental Station of Northwest Agriculture and Forestry University and analyzed the fruit color feature and growth character of Hayward cluster kiwifruit with scaffolding cultivation mode. Then overall plan of kiwifruit picking robot and fruit recognition method were determined. Identification target fruit from the bottom has advantages like simple background, less interference and so on. Through fruit image analysis, color image segmentation method 1.1R-G was adopted. After a series of image processing and feature extraction, image processing method of fruit pixel coordinates of feature points was gotten. Experiments show that the success rate of kiwi fruit recognition in this method reaches 89.5%.(2)Artificial light method. In order to meet the needs of Kiwi fruit picking robots all day work, method of using artificial light, LED, at night was studied LED. By collecting and thresholding fruit image under different light intensity, optimum light intensity was determined after contrast with hand-split image. According to the tests, it is found that when the light intensity between 30-50 lux, color image segmentation of 1.1R-G method works best at night. Under these conditions, identification of fruit is 92.2%.(3)Mathematical model of coordinate conversion. Coordinate conversion is the intermediate link of space coordinate transformation which links the space coordinate of image processing and Kinect sensor. Through analysis the relationship between Microsoft camera image and pixel coordinates of Kinect sensor image, mathematical model of coordinate conversion was built. Spatial coordinates of Kiwi fruit feature points were obtained based on the mathematical model. Tests show that the model can accurately perform a coordinate transformation.(4)Target fruit spatial coordinates acquisition. Target fruit spatial coordinates acquisition is the ultimate goal of the visual system. To accomplish that goal, first working scope and principle of Kinect sensor was analyzed; then Kinect sensor coordinate obtain program was written in Visual Studio 2010; finally spatial coordinate acquisition accuracy experiment was proceed which indicates that the error of spatial coordinates is less than 3mm.(5)Evaluation experiments of Kiwi fruit picking robots pick. Combine the research results and existing picking robot prototype, fruit picking experiments at daytime and nighttime was processed. The results of these two evaluation experiments show that the image recognition methods can accurately extract kiwifruit features and get pixel coordinates of fruit feature points; spatial coordinate acquisition method meets the operation requirements of picking robot; Artificial light method could satisfy the need of image recognition and robot working at night.