Machine Vision System for Autonomous Agricultural Vehicle

The problem of the vehicle motion control for automated driving is considered. The robotic chassis is designed for agriculture and should work in the absence of roads. The general structural diagram of the chassis' machine vision system is given. A structural obstacle separation method for constructing obstacle maps on the ground is proposed. The method uses lidars to detect obstacles. The implementation of the method is based on the assumption that the terrain within the area no significant differences in elevation (gullies, dips) and water obstacles. For solving the detecting obstacle problem in these conditions, it is enough to detect points which height exceeds a certain threshold, and to identify the relationship between these points to assess the obstacle size. The clustering of points in each layer by the Euclidean distance is performed. Then, the coordinates of the cluster centers are recalculated into the global coordinate system, which allows transferring obstacles to an area map, taking into account the dimensions that determine the degree of the detected obstacle danger. An algorithm for implementing the proposed method is described. The report also provides information on the composition of the software for the robotic chassis machine vision. Simulation results of the proposed method of the allocation of obstacles are presented. The method has low computational complexity, which reduces the requirements for the robotic chassis on-board computer.