Metric and appearance based visual SLAM for mobile robots

Simultaneous Localization and Mapping (SLAM) maintains autonomy for mobile robots and it has been studied extensively during the last two decades. It is the process of building the map of an unknown environment and determining the location of the robot using this map concurrently. Different kinds of sensors such as Global Positioning System (GPS), Inertial Measurement Unit (IMU), laser range finder and sonar are used for data acquisition in SLAM. In recent years, passive visual sensors are utilized in visual SLAM (vSLAM) problem because of their increasing ubiquity. This thesis is concerned with the metric and appearance-based vSLAM problems for mobile robots. From the point of view of metric-based vSLAM, a performance improvement technique is developed. Template matching based video stabilization and Harris corner detector are integrated. Extracting Harris corner features from stabilized video consistently increases the accuracy of the localization. Data coming from a video camera and odometry are fused in an Extended Kalman Filter (EKF) to determine the pose of the robot and build the map of the environment. Simulation results validate the performance improvement obtained by the proposed technique. Moreover, a visual perception system is proposed for appearance-based vSLAM and used for under vehicle classification. The proposed system consists of three main parts: monitoring, detection and classification. In the first part a new catadioptric camera system, where a perspective camera points downwards to a convex mirror mounted to the body of a mobile robot, is designed. Thanks to the catadioptric mirror the scenes against the camera optical axis direction can be viewed. In the second part speeded up robust features (SURF) are used to detect the hidden objects that are under vehicles. Fast appearance based mapping algorithm (FAB-MAP) is then exploited for the classification of the means of transportations in the third part. Experimental results show the feasibility of the proposed system. The proposed solution is implemented using a non-holonomic mobile robot. In the implementations the bottom of the tables in the laboratory are considered as the under vehicles. A database that includes di erent under vehicle images is used. All the algorithms are implemented in Microsoft Visual C++ and OpenCV 2.4.4