Information Fusion Technology Based On DSmT And Its Application In The Map Building For Robot

With the development of science and technology, the information fusion technology has become one of the hottest points in scientific researches. Because of its wide application prospect in both military and civilian fields, information fusion has been paid much attention to by the scholars in the world. Many theories and methods about information fusion have been presented, but they always have some limitations and their application is limited too. The fusion methods based on belief assignment have the advantages of measuring and combining the imperfect information. The Dezert-Smarandache theory (DSmT), which was proposed in recent years, has established a new approach in the information fusion fields. Here, the information fusion technology based on belief assignment, such as DSmT, has been studied deeply in this paper on the background of the map building for mobile robot in unknown environments.After the Dempster-Shafer theory (DST), the DSmT and the neutrosophic theory have been introduced, the multi-source information fusion based on belief assignment has been analyzed in this paper from different points of view, such as the quantitative information fusion and qualitative information fusion, the equally reliable information fusion and inequally reliable information fusion. To extend the fusion technology, the quantitative/qualitative neutrosophic belief assignment and combination rules have been proposed following the connection of DST, DSmT and neutrosophic theory.The methods processing stereo vision information always require much computation and have poor real time performance. To solve such problems, a quick interest point detecting method is presented in this paper. The method uses the scale-nomalized Laplacian operator to extract the interest points in the image pyramid and computes the feature vector based on the color information of the image area around the interest point. The experimental results have proved its validity. To extend the numbers of correspondence point pairs, the new interest point is used to lead the edge matching. On the basis of the method, the three-dimensional reconstruction based on the binocular stereo vision has been carried out to acquire the depth information, in order to attain the fusion of visual information and the fusion of the visual information, sonar information and laser information in map building for mobile robot in this paper.After the modeling of sonar, laser and vision, three experiments about 2D/3D grid map building for mobile robot, which are based on the DSmT fusion method, have been presented in this paper. One paid attention to the fusion of multi-source, congener and quantitative information, one put emphasis on the fusion of multi-source, heterogeneous and quantitative information and the other on the fusion of multi-source and qualitative information. Compared with the maps based on the other fusion methods, the DSmT has the advantage of fusing and processing the imperfect information. The experiments not only prove the validity of DSmT, but also provide a new way to the navigation and map building for mobile robot. Furthermore, a new two-level information fusion frame was also proposed in the paper to combine the multi-source heterogeneous information effectively.To conduct the experiments better, the Pioneer 2 mobile robot has assembled the binocular stereo vision system and the software experimental platform based on VC++ 6.0 and OpenGL has been designed. The control frame on two client-server levels, the module design and the excellent interface have rendered the software system to become an important tool in the experiments.