Vision-based Navigation Of Walking-type Rice Transplanter

As one member of precision agriculture, the rice transplanter's navigation is also very important. But so far more of the research is build upon riding-type transplanter, no material was found which shows the walking-type ones has been promoted. As the riding-type transplanter is also very useful in agriculture, so this paper has the following research about its vision-based navigation.1) Image segmentation between rice seeding and paddy field Various color space and color index (EXG, CIVE) were compared to find the best method for transformation of RGB color to grey levels; different thresholding methods were compared to find the best one for segmenting gray image into a binary one. CIVE color index with OTSU method was proved to be the most suitable combination. Global threshold with automatic threshold was combined also through weighting coefficients, for image segmentation, in order to reduce the influence of light intensity (e.g. strong reflex on the water surface). On the other hand, image erosion was selected to remove some noise and deduce the amount of data that the following algorithms have to process.2) Guidance line abstraction and guidance information extracting Hough transformation was selected for row tracking. But due to multiple lines would be extracted, so much more study was carried out on how to get the right one. Zhang's method is selected for camera calibration, and how to obtain camera parameters through OpenCV is also introduced. Navigation 3D model is set up which is used for deduce the computing formula about navigation parameters, such as distance and deflection between transplanter and seeding rows.3) Structure modification of transplanter for vision-based navigation An additional hydraulic system was designed for electric-hydraulic control; one extra swing wheel was added to achieve small angles turning of transplanter. Because of the different motion type between oil cylinder (straight line motion) and tail wheel (pendulum motion), motion analysis is taken in order to get the most suitable actuating mechanism.4) The experiments carried out for transplanter's vision-based navigation Experiments were carried out build upon the modified hardware and the designed software. Remote control about transplanter in both cement ground and paddy field was taken in order to get the navigation system's hardware performance. Vision-based navigation in cement ground was the preparation for paddy field experiment. The test results in cement ground showed an mean accuracy of 90 mm lateral error and 2.4°heading error. The standard deviation was about 65mm and 4°. With the test result of navigation system, the further modification has been promoted in order to achieve better performance.