Shape And Volume Estimation Of Kiwifruit Using RGD-D Camera

Kiwifruit is a fresh and delicious fruit.It has a high nutrition and dietary value.The sorting of domestic kiwifruit is mainly artificial,time-consuming and laborious,and the artificially-sorted kiwifruit is less competitive in the international market.In order to achieve intelligent sorting of kiwifruit,this thesis studied the estimation method of kiwifruit shape and volume.A kiwifruit shape and volume estimation method based on RGB-D camera was proposed.The main research contents and results are as follows:(1)An image acquisition system was established to take images based on RGB-D camera.RGB-D camera can not only acquire the two-dimensional information and color information,but also acquire the three-dimensional information of the image.Therefore,this thesis selected the Kinect V2 as an information acquisition camera.Light source played an important role in image acquisition quality and subsequent image processing.In the experiment,the light source was distributed around the camera and installed at the same height as the camera.(2)The estimation method of the shape and volume of kiwifruit and the method of image information collection were studied.The length,maximum diameter and minimum diameter of kiwifruit were estimated using caliper several times.The weight of each kiwifruit was estimated using an electronic scale.The volume of kiwifruit was estimated using the drainage method.This thesis analyzed the image of kiwifruit obtained at different mounting heights of the camera and concludes that the optimal range of camera mounting height was 650-750 mm.Therefore,the installation height of the camera in this thesis was set to 650 mm,700 mm and 750 mm.Finally,the kiwifruit image information of different installation heights of the camera was obtained through an image acquisition system.(3)The kiwifruit shape estimation method was studied based on RGB images.Firstly,RGB images were preprocessed.Secondly,the background image of kiwifruit was removed through the YUV color space.Then,median filtering eliminated noise interference in kiwifruit images.And then,using erosion,expansion,and open operations can enhance the image and make the clearer outline of the kiwifruit.Finally,the edge of kiwifruit was estimated using the Canny operator.The length and the maximum diameter of kiwifruit were estimated by the color information of the kiwifruit.Regression analysis showed that length R~2 values of kiwifruit were0.96,0.95 and 0.94,and RMSE were 4.61 mm,4.72 mm and 5.09 mm when the camera installation height was 650 mm,700 mm and 750 mm,respectively.The maximum diameter R~2values of kiwifruit were 0.91,0.91 and 0.87,respectively,and the RMSE were 5.12 mm,5.09mm and 5.27 mm,respectively.(4)The estimation method of shape of kiwifruit was studied based on point cloud image.When the camera installation height were 650 mm,700 mm and 750 mm,the length R~2 values of estimation were 0.96,0.96 and 0.95,respectively,and the RMSE were 4.78 mm,4.38 mm and 4.80 mm,respectively.The maximum diameter R~2 of estimation was 0.92,0.92 and 0.88,respectively,and the RMSE were 5.00 mm,4.74 mm and 5.21 mm,respectively.The estimation of the minimum diameter of kiwifruit is not affected by the absence of the edge region of the fruit image.The test results showed that the average absolute error of the minimum diameter of kiwifruit were 4.05 mm,3.99 mm and 4.14 mm,respectively.(5)The estimation of kiwifruit volume and its relationship with quality were studied based on the linear regression.The point cloud images and color images were used to estimated the volume of different varieties of kiwifruit at different heights.The results showed that with different installation heights of the camera,the estimation volume based on the point cloud image was higher in accuracy than the estimation volume based on the color image.At the same installation height,Hayward’s volume estimation accuracy was higher than Xu Xiang’s volume estimation accuracy.When the installation heights of the camera were 650 mm,700 mm,and750 mm,the R~2 values of the kiwifruit volume based on the point cloud image estimation were0.92,0.93,and 0.93 respectively,and the R~2 value of the kiwifruit volume based on the color image estimation was 0.90,0.90,0.91,respectively.A linear regression was performed based on the volume of kiwifruit estimated by the point cloud image and the quality of kiwifruit,the R~2 values were 0.96,0.96,and 0.95,respectively.This showed that there was a high correlation between the estimated volume and the quality of kiwifruit,so this method can be applied to the volume and weight estimation of kiwifruit.