The Outdoor Road Environment Based On Heterogeneous Vision To Create Topological Map

The mobile robot autonomous navigation technology is a hot topic of robtics research domain has a wide range of applications in both civilian and military.For autonomous navigation in an unknown environment, the mobile robot need to use its own sensors to build environmental maps. The map building is the basis for other tasks.Nowadays, many buliding map methods are presented for indoor environments.However, they less suitable for large-scale outdoor environments. Vision sensor can get a wealth of environmental information and panoramic vision camera has a360°viewing angle with a wide field-of-view, more and more used in autonomous navigation.In this paper/we propose a method of topological mapping which apply to outdoor road environment based on iso-vision system (omnidirectional vision and front vision).Firstly, this paper researches the principle of omnidirectional imaging system and related technologies, and compares several different types of omnidirectional cameras.we chose a hyperboloidal catadioptric omnidirectional camera and a common front vision camera to bulid iso-vision system.Then we research and propose a method for crossroad detection based on omnidirectional system. The method based on the result of omni-vision image segmentation. We proposed an improved region growing algorithm combined with threshold algorithm. The method uses the maximum entropy method to select the initial seed point for region growing, and integrated two different growth criteria to implement regional growth. The experimental results show that the method is very suitable for image segmentation of the panoramic vision.. We test three methods of detect crossroad based on the result of image segmentation, and carried out a lot of experiments.Finally, we chose a method which uses the robot location of the road the number of branches and branch direction to distinguish the road environment.Thirdly, we research the extraction and matching of natural landmarks based on front vision, and we study some improtant methods of image local feature extraction and detection.we implement a SURF detector-descriptor scheme with respect to robustness, yet can be computed and compared much faster. According to the requirement of this paper, we reduce the ocatve of the SURF pyramid, which is more real-time and saving the mapping time and space consumed.Finally, we propose a novel topological map building scheme based on previous research. Nodes are detected using the roadcross detection method based on omni-vison which is proposed by us. Nodes are represented with front visiual one octave SURF feature which include every road branch.The topological mapping process without a precise global location, and it is easy to build and maintain, and suitable for autonomous navigation in large-scale outdoor road environment without any landmarks.