A method to estimate robot's location using vision sensor for various type of mobile robots

The aim of our research is to realize a navigation method that can be applied to various type of mobile robots and environments. In this paper, we propose a method to estimate robot's location for the situation in which camera posture varies irregularly. Our method consists of two processes, generation of a feature vector from the image acquired by the robot and identification of a location of the robot using Support Vector Machine (SVM). The feature vector is generated from a histogram of image features extracted from an image divided into small regions. In each region, the average of color difference value and the number of edge pixels are calculated as an image feature which is not sensitive to small change of position and rotation of the camera. When the robot moves from a location to another location with acquiring images, the sequence of acquired images can be divided into segments based on similarity of feature vectors. Then we apply SVM to identify acquired images based on feature vectors, and the identifier of one segment to distinguish from other is generated based on SVM. Once the robot obtains identifiers, it can estimate its location based on the output of identifiers. To evaluate our proposed method, we experimented in a real environment with wheel type robot on uneven ground and a leg type robot.