Laser detection method for cotton orientation in robotic cotton picking
wang ling,liu siyao,lu wei,gu baoxing,zhu rongjie,zhu hongchao
DOI: https://doi.org/10.3969/j.issn.1002-6819.2014.14.006
2014-01-01
Abstract:In order to detect cotton’s position on a plant,a laser measurement experiment was designed,which included a crossgirder,and beneath which a model LMS291-S05 laser scanner produced by SICK Co.,Ltd was fixed. Then the three-dimensional coordinate values of a single cotton plant can be measured with the devices mentioned above,and the acquired distance image was furthermore processed with computer image processing technology and pattern recognition theory and also with the agronomic characteristics of cotton taken into consideration. The sampling interval of the laser sensor on axis X and axis Y was both 0.004 m,and the sampled points formed a image that illustrated the distance from the laser sensor to the plant surface point cloud. For the purpose of removing the unneeded background of the plant in the image,the acquired plant distance image was first processed with binaryzation at the threshold value of 0.9 m,because that 0.9 m happened to be the longest distance from the laser sensor to the bottom part of the cotton plant surface according to the measured data. Observing the cotton plant’s morphological characters in the formerly binarized image,and it was not hard to find out that the cotton part in the image appeared to be circular with a large area,while the stalk part in the image was thin and tiny. On this basis,morphology opening operations were carried out toward the edge of the bimarized image with the structural elements from a circle whose radius ranges from 0.01 m to 0.02 m,which is the usual width of a cotton stalk,in order to remove the unneeded cotton stalk background and extract the binary image of cotton. After a series of comparisons among different distances namely Euclidean,Mahalanobis,Bullock,and Minkowski distances,and considering that the pixel gray value was one-dimensional data,the distance between pixels was found to be insensitive to computing methods. The similarity between pixels in the image was calculated with Euclidean distance,and the centroid distance was chosen,which was also decided among several other different distances like closest distance,longest distance,and sums of squares of deviations,as the between-class distance according to the Cophenetic correlation coefficient. And due to the fact that branches that were 45° to each other usually at least had two more branches in between them vertically,and the minimum vertical distance of two neighboring branches was about 0.05 m,correspondingly the minimum vertical distance of two pieces of overlapped cotton was about 0.17 m,so the cotton image clustered at the threshold of 0.17 m. Then the adhesion or overlapped cotton was segmented,and the three-dimensional coordinate values of a single cotton plant were acquired. The result of a series of experiments showed that the recognition rate of a single cotton plant was as high as 96.67%,the mean measurement error of the cotton was 0.015 m,and the relative error was 2.43%. The correlation coefficient between the laser measurement and manual measurement results was up to 0.9934,which was high enough to provide the picking robot with needed parameters that could be used to determine its movement locus.