| With the background of the development trend of international agricultural engineering in 21st century, modern technologies of electronics, informal on and automation are introduced into traditional agriculture vehicles to realize visual navigation, which will become the base of agricultural machine possessing abilities to exert precision agriculture, so that labor can be reduced, pollution caused by agricultural chemicals can be avoided and agricultural cost can be decreased. The main and creative achievement include following aspects:1. The methods of detecting the road followed by agriculture wheeled mobile robot guided by machine vision are researched which are based on region segmentation and edge detection respectively. Appropriate characteristics to segment the image of agricultural field are analyzed, and 2G-R-B is considered to be able to provide the best result of segmentation. The holes in the binary image after segmenting are removed by "close" algorithm based on morphology. The directing function of knowledge to detect the guiding road is analyzed, and the line road model is provided according to the characteristics of the agricultural environment. The principle of multi-resolution dividing space based on wavelet transformation is introduced, and multi-resolution edge detection is presented by using B-spline wavelet. Results of processing many images of agricultural field and scene of navigation experiments prove that the method based on region segmentation can detect the road correctly, the multi-resolution road detection based on wavelet transformation can easily analyze the image on different scaling, and can possess the power to get main contours while suppressing small edges when the scale is large. The latter, however, is found to be unable to satisfy the real-time requirement of navigation due to the complicated calculation.2. Generally, traditional Hough transformation is used to detect the line or some special curves in the image space, and navigation parameters can only be obtained indirectly by finding the corresponding position in the navigation coordinate through the anti-perspective matrix. The motion law of the perspective projection of a line on theground in the image space is analyzed, and the navigation parameters are gotten directly by transforming the line in the image space into navigation parameters space according to the principle of Hough transformation. Three group experiments are conducted under the different conditions, and the results validate the analysis about getting navigation parameters directly. The measuring precision of lateral and heading displacement is similar to the level reached in the same kind of international researches, with the standard error of 3cm and 0.62 degree respectively. The orientation of selecting some p; rameters is clearer by comparing; the results of three group experiments when utilizing visual navigation in the agricultural field. This algorithm will be able to improve the real-time of detecting road, overcome the road loss caused by shadow or missing plant and conquer many other disturbances.3. The behavior characteristic of the agriculture wheeled mobile robot is analyzed, and the corresponding state equation and measuring equation are deduced. Including the testing of the validity of sensor data and the adapting of the Q matrix in reai-time, the Kalman filter is researched to improve the real-time, precision and robust of navigation system. With this method, the disturbances resulted from the many uncertainties in non-structure agricultural environment can be overcome, the movement of road line can be followed in the image space so that the technology of visual window can be used, the current measured state of wheeled mobile robot can be substituted by the predicted state to increase the real-time of control system. Both the simulation and the experiment show the effectiveness of this method.4.The nested control loop is analyzed, and the bang-bang control is used in the inside loop while the direct l...
|
PDF Full Text Request