Research On The Vision Navigation Information For Autonomous Mobile Agricultural Robot

The study aims at providing the navigation information for the autonomous mobile agricultural robot. On the basis of fully generalization and inheritance of advanced research home and abroad, the new image process method is explored and applied. Through the process of cropland image information of complicated environment, the agricultural robot can attain real environment information with CCD camera. The targeted path can be extracted and the relative pose between robot and path is determined, It lays foundation for visual navigation and moving to the desired target without the interference of man. Cropland image analysis is the basis part of information process technology of visual navigation for the autonomous agricultural robot. Owing to the influence of factors such as lighting, image noise, angle of portray and distance, the object in image such as ditch, ridge lose most of their detail contour and the edge is not satisfied, which display the texture property in macro. The traditional methods of differential opera to extract regional character are not effective. Automatic detection of the cropland areas under the complicated background is dealt with in the chapter 2. Some knowledge-based approaches are introduced to implement the area detection. In order to detect areas, a lowpass filter is used to preprocess the original images at first, then binary image is obtained by means of grad transformation. After the thin texture is moved with mathematical morphology thresholding dilation, the complete edge of cropland can be obtained. All of the algorithms proposed herein are knowledge-based. Even the binary images of cropland are occluded or fragmentized, the areas can be detected efficiently and effectively. Experiments show that this method is reliable and effective. Vision navigation and location is a main function of the vision system of the intelligent agricultural mobile robot, and the edge feature of images is an important feature for vision navigation and location, which can affect the 3 accuracy of scene matching and location when navigating. In chapter 3, a set of filters are given that can be used to construct the edge detecting wavelet for edge localization after researching the edge detection of images with wavelet. According to the characteristic of cropland scenery, a compactly supported dyadic and antisymmetric with respect to origin wavelet is selected to detect edges of cropland image. The image edge features are provided by determining the local maxim of wavé‘œet coefficier4t at dvadic scale of the image. After the computer simulation is carried out on the cropland image, the continuous and smooth edge image can be attained. The edges of cropland scenery are extracted accurately The experiment results indicate that the method is efficient and practicable. Vision location is a main function of the \iSiOfl system of the agricultural robot. To fully automate some agricultural tasks in field, an agricultural robot has to be guided along a cropland edge. A method of the agricultural robot vision locative is put forward. The method is based on extraction of the scene structure from pcrspective information that is a set of coplanar parallel lines representing the cropland edges. To predict future course with the information in the visual domain, the image geometry of the agricultural robot mold model is constructed. A vanishing point detection algorithm is pe...