Research Of Monocular Vision-based Obstacle Recognition Technology For Unmanned Ground Vehicle

Unmanned ground vehicle has broad application prospects in many areas, such as military, civilian and scientific research, et al. The obstacle detection and recognition technology of its key technologies has been a hot research, but also difficulty, especially in off-road environment. This paper research the obstacle detection and recognition technology in the off-road environment based on monocular vision. The types of obstacles researched in this paper are trunk, high shrubs and dwarf shrubs. The main contents are as follows:In the process of image acquisition based on monocular vision, there will be some noise, so median filtering algorithm is used to filter out the noise. This algorithm can filter out noise, and can maintain the edge characteristics of the image, which is in line with the requirement of needing edge information. In addition, the improved segmentation method based on Fisher criterion and K-means clustering are used to extract the region of interest. The test shows that the method has good segmentation results for obstacles, such as trunks and shrubs.In order to extract the obstacle feature used to obstacle recognition algorithm, morphological processing method is used to further process the segmented image, and it can very effective to remove the interference generated by over-segmentation. Then, this paper extracts the feature, S-feature and a-feature, based on HSI color space and Lab color space. And chain code is used to extract the feature of verticality in the binary image. That laid a good foundation for obstacle identification.Based on Bayas classification principle, this paper analyzed a large number of training samples to estimate the parameters of Bayesian classifier based on a single characteristic. Then this paper presented multi-feature fusion algorithm based on Bayesian classification principle, and the obstacle recognition results based on a single feature and multi-feature fusion can be obtained through using MATLAB test platform designed in this paper. Finally, a lot of tests and comparative analysis verify the proposed multi-feature fusion algorithm has good usability and accuracy.