Obstacle detection by pattern matching of image templates based on the self-organizing map

Abstract

This paper presents a method of obstacle detection using a SOM-based template matching approach. The SOM is the clustering technique with the training function. By the training of the SOM, the resemble images with several images taken from the working environment of the robot are generated, where the resulting image database is called the template image set. The detection of the obstacles is implemented within a preliminary divided region in front of the camera on the robot. On the obstacle detection, the query image taken from the environment is compared with the template images using a simple matching algorithm in order to determine the template image associated with it. By carrying out the template matching, it is clarified that the presence and the location of an obstacle can be determined within an accuracy in a use of a mobile robot. Our approach is tested by a number of experimentations in an indoor environment setting.