| The research of agricultural machinery navigation path based on machine vision is mainly concentrated on drought crop row, ridge, furrow and harvest border and so on. It is rarely applied on paddy field tillage machinery. The experimental study on the method of path recognition and parameter extraction of high stubble paddyfield-tillage based on machine vision navigation was conducted. To achieve the objective, the paper draws lessons from existing path identification and parameters extraction theory and method to assess rice, rape and wheat high stubble paddyfield-tillage images which are obtained based on1GMD-200paddyfield crop returning cultivator. Main research contents of this paper are conducted as follows:(1) Build up dynamic acquisition system of high crop paddyfield-tillage image. The high stubble paddyfield-tillage images are collected dynamically by industrial camera controlled by computer on the platform of eastern tractor LX-954. The mounting height of camera above ground is2.15m and angle of depression is31.8(2) Propose a new gray-weighting factor R+G—2B and obtain a better graying effect. Compared the graying results of normal weighted method, excess green weighted method and hue value method, it concludes that using excess green weighted method can gain contrasting gray image. In order to search for a better graying effect, color component characteristics in non-tilled and tilled areas of high stubble paddyfield-tillage images are studied and propose a new gray-weighting factor R+G—2B.(3) Put forward an algorithm for image segmentation based on pixel mean value of texture descriptors. Through comparing the segmentation results of excess green gray image and R+G—2B gray image by gray histogram arithmetic, OTSU method and texture statistical approaches, the method for image segmentation of this study is pixel mean value of texture descriptors based on R+G—2Bgray image. The window size of pixel mean value of texture descriptors is12by2in pixels and the threshold value is100.(4) Present a cropping binary image algorithm for large area of white points or block in tilled area. The cropping binary image could be obtained like this:first, compute accumulative pixel values of the binary image in vertical direction; then, figure up the mean value A of the maximum and the minimum of accumulative pixel values; last, find out the pixel value1.8×A and0.1×A corresponding the number of columns.(5) Determine the least square method to be navigation path detection method in this study. Navigation path detection of high stubble paddyfield-tillage images has been carried out by the methods of the least square, the traditional Hough transform and progressive Hough transform. Results indicated that using the least square fitting navigation path is high precision and strong anti-interference, with time-consuming600ms; using progressive Hough transform is higher precision and less time consuming than the traditional Hough transform, but both are easy to detect navigation path mistakenly. They take1500ms and450ms respectively.(6) Calibrate camera intrinsic parameters. Zhang Zhengyou plane calibration method is used to calibrate the camera. Road error verification test shows that the error of tilt is within1and the error of intercept is in the range of0-0.03m. Calibration results can meet the actual needs of navigation.(7) The test of tracing road signs on the road in campus show that the intercept error is Δc=0.3256mm and tilt error is Δ(?)=2.4156. The field experiments in rice core test area of modern agricultural science and technology in Huazhong agricultural university show that the intercept error is Δc=-0.1655mm and tilt error is Δ(?)=-3.3729°. |
PDF Full Text Request