Recognition And Location Of Green Apple Based On Binocular Stereovision In Night Environment

Apple planting has a strong seasonality,the work is basically completed by manual in the process of picking,the working time is concentrated,the labor intensity is high and the efficiency of picking is low.The application of the technology of picking robot can reduce the workload of manual picking.Machine's vision system is an important part of apple picking robot.At present,there are few research on the recognition and location of green apples.This paper takes the night green apples as the research object,and focuses on image denoising,image enhancement and recognition and location,which solves the problems of complex illumination and partial occlusion encountered in the process of recognition and positioning and realizes the night operation mode of picking robot.The main research contents and conclusions are as follows:(1)The appropriate auxiliary light source is chosen and the noise characteristics of the night green apple image is studied.The characteristics of the median filtering algorithm and the wavelet denoising algorithm are analyzed.Combining the advantages of the two algorithms,the denoising of the night green apple image is processed and compared with the common denoising algorithms.Combine the advantages of to denoise the night green apple image and compare them with the common denoising algorithms.Taking PSNR as the evaluation criterion of image denoising,the PSNR of the denoising algorithm in this paper is the highest,that is,the image contains the least noise.(2)Aiming at the problem of global darkness and loss of detail information in night green apple image,an image enhancement algorithm is proposed to protect detail features.On the basis of Retinex image enhancement theory,the original image is processed by using improved LoG operator,and the detail features of the image are extracted.According to the image information entropy,the proportion coefficient is calculated.The detail features are integrated into the image processed by single-scale Retinex algorithm,and the enhanced result image is obtained.Compared with common image enhancement algorithms,the image enhancement algorithm in this paper can achieve good image enhancement effect and make images to carry more detailed features.(3)Based on the idea of template matching,the recognition algorithm of nightgreen apples is studied.The standard green apple boundary is obtained by using the shape characteristics of mature green apples.Through seven scales transformation,a template library is created to meet the actual requirements of picking.NCC template matching algorithm is selected to match and image pyramid algorithm is used to accelerate the process of matching.Repeated experiments show that the matching accuracy is 90% without occlusion,and the successful matching can be achieved when the area of occlusion is less than 40%.The average time of matching a template image is 0.8 seconds.(4)The binocular stereo vision system of green apple at night is studied.The left camera is defined as the origin of the world coordinate system;the camera calibration is completed with the help of Matlab platform;the inside and outside parameters of the binocular camera are obtained and the binocular camera is corrected.According to the principle of binocular vision,the location of the picking point of green apple is calculated.Experiments show that the errors generated in the distance between the left camera and the green apple from 500 mm to 1500 mm can satisfy the requirements of actual picking.