On-line Recognition And Location Of Camellia Oleifera Fruit Based On Stereo Vision

Camellia oleifera is one of the four famous oil source tree species in the world.At present,picking Camellia oleifera fruit mainly depends on manual picking,which is a very great amount of labour and extremely inefficient.Camellia oleifera fruit picking robot can reduce labor intensity of workers,improve labor productivity,and ensure picking timely.How to identify and locate Camellia oleifera fruit in picking process is the key and the difficult point of Camellia oleifera fruit picking robot.This article unifies domestic and foreign visual picking research experience to obtain images of camellia oleifera fruit under natural environment,and proposes an online recognition and location technology of camellia oleifera fruit based on stereo vision,which provides a visual basis for Camellia oleifera fruit picking robot.The specific research content is:(1)On the basis of analyzing fruit picking and stereo vision technology at home and abroad,analyzing the recognition and positioning environment of camellia fruit,choose Intel Real Sense D435 i as the depth acquisition device to build a stereo vision system to acquire images of camellia fruit under natural conditions.Use Dynamic Calibrator to calibrate the depth camera and color camera to obtain the internal and external parameters of the camera.At the same time,taking manipulator picking as the application goal,the composition of the camellia fruit visual system and the process of visual recognition and positioning of the camellia fruit are proposed.(2)Combining the identification and positioning process of Camellia oleifera fruit,construct a Camellia oleifera fruit data set,and use image segmentation and bicubic interpolation to optimize the data set.The Mask R-CNN network is used to segment the camellia fruit in natural picking environment,and the Inception＿v2 is used as the Mask RCNN framework of the feature extraction network for training,and then the target is recognized and segmented.At the same time,the same batch of training samples are used for target recognition in the SSD network,and the Snakes algorithm is used to segment the change area,and the running speed and recognition accuracy of the two methods are compared.The experiment found that the Mask R-CNN network with Inception＿v2 network as the feature extraction network has an AP value of 0.838 for the optimal method of recognition,and the average detection time is 0.892 s.(3)Aiming at the identified Camellia oleifera fruit,the least squares ellipse transformation is used to restore the contour of Camellia oleifera fruit to obtain the center coordinates and parameters of the camellia fruit.Align the RGB image with the depth image.At the same time,in order to avoid information scarcity in the depth image,the Telea algorithm based on the error model is used to denoise and repair the depth image.Taking the center coordinates of Camellia oleifera fruit as the center of the circle,obtain a circular area occupying 10% of the pixel area of Camellia oleifera fruit,and use the average depth of the circle in the improved depth image as the target depth value.(4)This paper proposes a fusion algorithm of Camellia oleifera fruit coordinates and depth information to identify and locate Camellia oleifera fruit.A human-computer interaction interface is designed to display the online detection results,and the effectiveness of the depth image restoration algorithm is verified through the depth positioning error test,and Camellia oleifera fruit recognition method and depth information are further combined with Camellia oleifera fruit identification and positioning fusion test.The correct recognition rate of Camellia oleifera fruit is 83.0%,the average time is 0.91 s,the positioning error is 6.1mm,and the average length and short diameter deviations are 3.4mm and 3.0mm.