Using image processing methods for a radar estimate of marine vehicle odometry

Abstract

Independent navigation verification is an important function in navigation. The current state of the art is to manually plot radar ranges or visual lines of bearing to landmarks that are distinguishable both on the sensor being used and on the chart--this requires highly trained personnel. In automated navigation this problem can be solved with a Simultaneous Localization and Mapping (SLAM) process using an outward-looking sensor, like a LIDAR or a RADAR. One of the inputs for a SLAM system is some form of odometry. This thesis looks at implementing real-time odometry and positioning in a marine environment, based on a single radar sensor. The radar used is widely available on ships, and this system could be implemented as a non-intrusive, real-time odometry measurement. Unlike other methods for odometry, like dopper logs, dead reakoning, or shaft logs, this radar method is able to estimate the impact of current and wind on the vessel's movement. Different configurations of the algorithm were used to try and minimize the impact of radar noise and vehicle rotations. The final method is able to run in real-time on a vehicle and accumulates about 1.5% error, at-least in inland waters. This demonstration was conducted off-vehicle on real-world data that was collected and then fed into the solver sequentially.