Research On Vision Based Localization Using Epipolar Geometry

Nowadays, with the development of social economy and the popularity of smart phones as well as intelligent mobile equipment, the service of indoor localization has been taken more seriously. As for the traditional positioning signal source, it is difficult or expensive to be acquired in the small scale indoor environment. In order to solve this problem, vision based indoor localization system has become one of the most promising development directions. Given that visual information can be acquired flexibly even in the complicated small scale indoor environment, visual information of indoor scenario can be utilized to carry out positioning mission in the indoor localization system.This paper analyzes the research purpose of vision based localization and expounds current techniques about indoor localization and vision based localization in our country and other countries. Besides, some related theoretical knowledge of image retrieval and epipolar geometry constraint in the vision localization system have been researched in this paper. Based on these work, this paper has finished the following research work:(1) Given that the time consuming of traditional method of building vision based indoor localization database, which is difficult to meet the need of building database or updating database immediately. Aiming at this issue, a fast database building method which utilizes video streams has been proposed in this paper. This method can decrease the complexity of database building and updating process dramatically.(2) Aiming at the issue of huge time consuming in the image retrieval process of existing vision based indoor localization system, after building the database in a fast method, based on the characteristics of this database and the need of vision based localization system, a clustering based fast image retrieval algorithm has been proposed. Besides, coarse matching combined with fine matching has been used to modify the existing image retrieval process in the vision based indoor localization system. The simulation results indicate that if the number of image in the database is huge, the clustering based fast image retrieval algorithm can decrease the complexity of image retrieval process in the vision based indoor localization system.(3) Traditional eight points algorithm is not robust enough, which leads to the descending of the accuracy of indoor localization. Aimi ng at this particular issue, a blocking based modified eight points algorithm has been proposed to calculate fundamental matrix. In the process of calculating fundamental matrix, traditional eight points algorithm might choose all eight points in a concent rated way, which would increase the error of fundamental matrix as well as indoor localization. However, blocking based modified eight points algorithm can avoid the above concentrated scenario and enhance the robust of fundamental matrix. Consequently, adopting blocking based modified eight points algorithm can decrease the error of fundamental matrix as well as indoor localization. The simulation results indicate that adopting blocking based modified eight points algorithm in the epipolar geometry based indoor localization system to calculate fundamental matrix can improve the accuracy performance and decrease the average error of indoor localization.