|Mir Alireza Athari|
Online simultaneous localization and 3D mapping based on point cloud representation in mobile robots
The main goal of this thesis is to design a simultaneous localization and 3D mapping based on points cloud representation system for a mobile robot. The sensor used in this system is a RGB-D camera. In the following thesis, Front-end & Back-end of this system is described and in front-end section FOVIS library is used for visual odometry and by adding depth information to feature matching section of FOVIS, the error of motion estimation is decreased. Next, a hash function called Radial Variance based hash is used for loop detection and for improving the accuracy, images are compared against the map. In the Back-end we have used iSAM2 framework alongside with switchable constraints robust loop closing method. In the robust method we have discussed about the effects of initial weights and an initialization method is provided which improves the results. Finally the whole SLAM system is implemented on a mobile robot from K.N.Toosi University called Avril and the verification results are provided.