Applications of stereoscopic vision to agriculture

Machine vision is a technique to sense reality by capturing an image with a camera and processing it with a computer to extract useful information. When one single camera is used to acquire images, the result is a flat, two-dimensional representation of the scene. On the other hand, if two similar cameras arranged to possess stereo properties are employed, the third dimension, depth, is available, with the consequent improvement in visual perception. The research presented in this document introduces a methodology to unite stereoscopic vision and agricultural applications.; A system to take stereo images and process them was developed. Images were grabbed from both ground and aerial platforms. Some of the applications described here required off-line processing; others were prepared for a real-time response. The agricultural applications pursued with this stereo system were: generation of 3D crop maps, obstacle detection, crop structure perception, and autonomous navigation for off-road vehicles.; Spatial maps generated via GIS (Geographic Information Systems) typically provide two-dimensional information. A stereo application was developed to render three-dimensional maps of crops and orchards. Images were acquired by a compact stereo camera, stored by a computer and analyzed by a developed algorithm. Global information was granted by a GPS (Global Positioning System) receiver. Coordinate transformations and map composition tasks were realized off-line by specific algorithms. Noise reduction, coordinate accuracy and individual images assemblage turned out to be crucial issues in the creation of a 3D field map. Different scenes were studied and several 3D maps were generated following the techniques developed for this research.; Two applications of stereovision for automatic guidance of agricultural vehicles were conceived. The first one employed disparity images together with regression analysis to find the vehicle's directrix. Problems appeared with poor correlated lines and vertical lines. A modification in the regression equations of vertical lines was incorporated. An algorithm to determine the target point within the disparity map was elaborated. This application was tested with real images obtained in the field. The second application was developed to a conceptual level. It utilizes three-dimensional images to estimate a vehicle's trajectory.