Path Planning For Autonomous Harvesting Robot Based On Stereo Vision And SBL-PRM

In recent years, the research of the autonomous harvesting robot has already become the popular issue, because of the shortage of the agriculture labor force both in developed countries and developing countries. Autonomous harvesting robot must apperceive surrounding environment information first, and then make integrated decision in accordance with the surrounding environment to make a reasonable plan for their moving, that is to say, path planning problem. Different from industrial robot, which works in particular environment, the autonomous harvesting robot mainly works in the natural environment. The autonomous harvesting robot has to face more complicated and uncertain circumstance; therefore there are more problems to be resolved.This paper focuses on key techniques of autonomous harvesting robot, such as environment perception based on binocular stereo vision, path planning and main works are accomplished as following:1. In order to realize the environment perception of autonomous harvesting robot, the dissertation studied the the recognition of obstacle and location technology. Previous algorithms were analyzed and a branches image recognition algorithm based on support vector machine was presented. After the branches region was segmented, the algorithm based on Normalized Cross Correlation was adopted for stereo matching. The algorithm, which is effective in practice, is simple, saving time, and insensitive to illumination. After stereo matching, combined with multi-segment approximation was used to calculate point's 3-D coordinates, and display the image drawn through OpenGL at last.2. A single-query bi-directional probabilistic roadmap planner with lazy collision checking was adopted for citrus picking robot real-time path planning in dynamic and unstructured environments. The conversion from the camera coordinate system to the world coordinate system was developed for robot localization. At the sampling and searching stage, the density of samples in neighborhood was used as weight to control expansion of roadmap and avoiding over sampling in free space. After finding out best path, we used hierarchical decomposition methods to detect collision, which based on oriented bounding boxes.Simulation of the planner was carried out in two cases of picking exposed and overlapped fruits. The effects of maximum number of milestones S, neighboring thresholdρ, local path checking thresholdεand path smoother steps N on average time and success rate of planning were analyzed. The simulation experiment shows that SBL-PRM is effective in the autonomous harvesting robot real-time path planning.Our research has made great progress in identification of obstacle, location and. three-dimensional reconstruction. The research results of this research have reference value for the study on visual recognition in the field of autonomous harvest robot in our country.