Study On Unstructured Road Recognition And Obstacle Detection Based On Machine Vision

With the rapid development of computer technology, intelligent agricultural machine will be one of the development directions of agricultural machinery. In the agricultural machinery operating environment, a large proportion of road is unstructured. The image processing technology will be used for road recognition and obstacle detection on the unstructured road. The researchers in domestic and abroad pay more and more attention on agricultural robot navigation technology based on machine vision technology. Many experts in the field of agriculture believe that machine vision technology is one of the essential technical measures that achieve agricultural modernization and sustainable development.Visual navigational problems for unstructured road conditions are systematically researched in this paper. The research studying on unstructured road identification and obstacle detection for agricultural robot based on machine vision is divided into three parts: image preprocessing, road edge detection and obstacle detection. The main contents of this research are as follows:According to the characteristics of unstructured road, HSV color space that corresponds with human visual perception is chose for analysis after comparing experimental analysis. A comprehensive analysis for a variety of image pre-processing technologies is made which including image filtering, threshold segmentation, region growing, providing high-quality image for edge extraction and obstacle detection on the road. The gradient magnitude algorithm and the method of Otsu threshold segmentation are used for the image segmentation of road area. The edge points of road is extracted initially through morphological algorithm and gray value differences between the road surface and the background. The experimental results still has good robustness in a complex environment. Based on Hough detection to optimize road edge points, RANSAC algorithm and least squares algorithm is used to determine the optimal random sample set of points. Finally, random sample points are curved to obtain the edge and achieve edge extraction of unstructured road.This paper presents an obstacle detection algorithm based on monocular vision. In view of the complex environment along the unstructured road and the changing image of the background at any time, two kinds of algorithms are used to analyze and solve problems.(1) The detection window technology. The frame of ROI detection window is set up in the forward area of the car. Obstacles are detected by adaptive region growing in the range of ROI. It reduces the search range of obstacles and increases the system speed at the same time.(2)The detection algorithm of moving obstacle around road edges based on optical flow method. The optical flow method and statistical theory are combined to estimate the current optical flow vectors of background. This estimation method can efficiently detect the target of moving obstacles under the condition of moving background.The algorithms of unstructured road recognition and obstacle detection in this paper provide a preliminary efficient way for autonomous navigation in the natural environment.