Research On Apple Recognition And Location Based On Binocular Vision

With the progress of technology and the development of society,the agriculture of our country is innovating as well.It is gradually shifting from traditional artificial farming to agricultural mechanization,agricultural automation and agricultural intelligentization.Apple is an important crop in China where is a big country of fruit planting.Fruit industry and fruit post-processing industries have become the core position in some regions.Marketing apple plant is an effective method to increase agricultural production and farmers’ income.At present,ageing of the population has become one of serious problem in China.While entering aging society,the gradual loss of the young and middle-aged labour and the lack of agricultural resources is becoming are undoubtedly becoming the major challenge to the development of agriculture.Therefore,it has intensely need to research on picking-robot development.It should obtain the information from external environment and targets firstly while the picking-robot is working.According to identify the target through image processing technology,the target will be located through localisation technology.This paper focuses on the apple-picking robot’s visual system,the detailed works are introduced as follow.(1)A visual platform is set up to provide a foundation for obtaining 3D information of the target fruit by the subsequent realization of binocular camera calibration,image correction,stereo matching and parallax operation.The visual platform includes a binocular camera,a bracket,a level and an adjustable shaft,and a computer.The program is written by Anaconda,Open CV,Matlab and Python.(2)The image processing method of the Fuji apple is studied.Firstly,comparing three kinds of colour space channel,the most suitable colour has been chosen to segment R-G channel.Secondly,Otsu algorithm and genetic algorithm for image threshold are combined for threshold segmentation.The optimal morphological operation is applied for processing the target fruit.After edge profile extraction by Canny operator,the feature extraction has been completed for targets fruit recognition according to optimise accumulator values from the circular Hough transform.(3)The camera imaging coordinate system and imaging model have been introduced,the principle of binocular vision and the Zhang calibration method have been studied as well.In order to obtain the information of binocular camera,the visual platform,Matlab,Open CV,and Python have been built for calibration test.Finally,the apple centre has been stereo matched that the 3D information of the apple has been obtained based on the principle of 3D reconstruction.