Research On Landmarks-based Vision Indoor Localization Algorithm

In recent years, with the significant development of the personal terminal's processing speed, there is a big strengthen in huge pixes image process, so vision-based indoor localization has become a hot area of research. Compared with the traditional algorithm, this method can be deployed with lower cost. In addition, it provides more robust localization results and extra visualized services.The localization results of the traditional method relies heavily on the density of position fingerprinting. Location accuracy can be improved when the fingerprinting is concentrated. However, this causes a greater time delay because of the bigger database and vice versa. This paper propose a vision-based indoor localization algorithm based on landmarks to solve above problems. It reduces time complex degree and improves the location accuracy by adding extra homography and projection constrain information.Firstly, this paper expounds the research objective and current techniques. Meanwhile, relevant theories of indoor visual localization algorithm are studied, including image feature extraction algorithm, homography constrain of the images and camera imaging model in space. This paper mainly completes the following studies:(1) A solution based on image matching and homography constrain is proposed for the unknown calibration line in user's input images, and it solves the equation automatically in online stage. By sampling and calibrating the calibration line of the landmarks in the database, this algorithm obtains the equation with SURF matching points and homography constrain between offline images and online images.(2) As for the problem that algorithm time complexity and location accuracy relay too much on position fingerprinting density, this paper proposes a combined localization algorithm based on several calibration lines in the landmark, and it obtains lower time complex degree and higher location accuracy. It uses several landmarks instead of redundant images in database. Moreover, additional priori information such as homography constrain between offline and online images, calibration lines' projection relations from landmark to image, to optimize the location results.(3) The simulation analysis of the calibration line solving algorithm and combined algorithm based on multiple calibration lines is presented. The results show that the calibration line solving algorithm can accurately obtains its equations. What's more, the combined localization algorithm improves the location accuracy, reduces the algorithm time complex degree, and obtains the user's direction angel as well.