Vi Congresso Nacional De Engenharia Mecânica Vi National Congress of Mechanical Engineering

Computer vision, when used in open and unstructured environments as in the inspection of crops, requires the use of specific vision systems acquiring algorithms prepared for such situations. These algorithms work mainly with images composed of complex objects, textures, shadows and brightness. The present project aims to apply an omnidirectional vision system for mobile robot navigation in agricultural area. Theses systems make the acquisition of image within a 360° field of view practicable. The arrangement of convex hyperbolic mirrors with lenses makes a computer vision system structure with larger vertical field of vision. Two methods of image correction were employed: Polar-to-Cartesian transformation and Direct hyperbolic transformation. Therefore, a panoramic image preprocessing module is deployed for edge detection, segmentation, and binary imaging algorithms, with artificial neural networks techniques, resulting in manageable images for space navigation and planting area.