Research On Vision Navigation Guidence Route And Navigation Parameters Acquisition Of Weeding Robot For Paddy Fields

The research of vision navigation guidence route and navigation parameters acquisition ofweeding robot for paddy fields is studied in this paper, including dectecting centerlines of riceseedlings rows and the realization of robot’s self-localization. One of the centerlines was usedas the robot’s navigation route. And it was used to calculate the navigation distance andnavigation angle. The concrete contents and conclusions are as follows.Since the paddy field environment varies from the different growth stages of rice seedlingand the color of rice seedling is similar to cyanophytes and duckweed contained in the paddyfield image in South China, the segmentation of rice seedling and centerlines detectionbecome very challenging. This paper presents a method to detect rice seedlings rows’centerlines based on color model and Nearest Neighbor Clustering Algorithm. Firstly,2G-R-Bcolor model, G-R color model, ExG-ExR color model and S component in HSI space wereused to extract the features of rice seedlings under different conditions. After comparation ofthe four methods, the most effective2G-R-B color model and S component in HSI space wereselected. Secondly, the Otsu or K-Means was used to binarize the gray image. Since thepunctiform noise and intersecting rice seedling’s leaves contained between rows in thebinarized image, morphology and statical scanning window alogrithm were used to eliminatethem. The feature points of rice seedlings were then extracted without changing their shapesand clustered based on Nearest Neighbor Clustering Algorithm according to their adjacentrelationship. Finally, the centerlines could be correctly detected by applying a known pointHough transform in each cluster. And the navigation distance and navigation angle werecomputed based on the centerlines.The experimental result showed that the feature of rice seedling can be precisely extractedunder a cyanophytes-and-duckweed-contained background, and the average processing timeof a320240-resolution color image is20~30fps. The average error of navigation distanceand navigation angle was4.1mm and1.1°respectively. The algorithms can be adapt to theenviroment variation and meet the real-time and precision requirement of robot.