A New Method of Global Image Analysis and Its Application in Understanding Road Scenes

Abstract

In this paper, real-time algorithms for constructing the adjacency graph and the spatial geometric relation between contrast objects of color images are proposed, as well as methods of global image analysis based on them, within the scope of the theory developed by the author. Image analysis is conducted based on the graph of color bunches STG and the bipartite graph LRG of left and right contrast boundary curves (germs of contrast global objects) in STG, introduced by the author. An essential point is that in each layer of this graph a linearly ordered covering constituted of “basic” color bunches is selected. Based on this covering, a search lattice for solving global problems of image analysis is constructed. The obtained results are applied to finding complex objects in images. In particular, they are applied to the analysis of road scenes. The developed methods are implemented in the form of a program complex. The results of its operation on video sequences taken from a moving vehicle are presented and discussed. The application of the developed technique to the navigation of autonomous robots is also considered.