Localization using single camera and Lidar

The location estimation problem, which is one of the basic elements of the navigation solutions, is mostly solved by using the global positioning system (GPS) and the data obtained from other sources (inertial measurement unit, digital terrain elevation datamaps, barometric pressure sensor, wheel odometer etc.). In GPS-denied environments,the position estimation problem is still an open problem in the literature. Any vehicle/robot(whether manned or unmanned, autonomous or not) must be able to perform accurate and high-precision location estimation in order to perform its navigation task even in GPS-denied environments. When the various methods used for the six degrees-of-freedom location estimation and the sensory equipment are examined in detail, it is understood that each has various advantages and disadvantages. It is obvious that the methods which use the data obtained from different sources together give better results in the solution of six degrees of freedom location estimation problem.In this thesis, six degrees-of-freedom (6-DOF) position estimation problem in GPS-denied environments is examined by using a single camera and lidar sensor. It is known that the position information which is obtained from a single camera is accurate up to an unknown scale factor. In order to estimate location of a vehicle with accurate metric measurements, this unknown scale information should be calculated by various methods. In this study, six degrees of freedom position information is calculated within the absolute scales by using the metric data obtained from the Lidar sensor with the pixels on the picture frames taken from the camera. Tests of the proposed algorithm were performed on the KITTI odometry dataset. KITTI odometry dataset were created with data taken from stereo cameras, Lidar, inertial measurement unit and GPS receivers which are placed on a moving car. A single color camera frames and Lidar sensor data in the KITTI odometry dataset were used to test the proposed algorithm in this thesis. Data from GPS and inertial measurement unit (IMU) were used as a reference to compare the success of the proposed Visual-Lidar Odometry algorithm. The test results obtained using the KITTI odometry dataset showed that the Visual-Lidar Odometry method, developed using a single camera data and Lidar scans, can produce consistent location information without the need for costly optimization methods.