Design of a Navigation System Based on Scene Matching and Software in the Loop Simulation

Abstract

This paper presents a design and a comparative study between two types of navigation systems based on scene matching. The first one depends on the cross-correlation and the other depends on Scale-Invariant Feature Transform (SIFT). A simulation environment that depends on Robot Operating System (ROS) was adopted with the 3D dynamic simulator Gazebo. The IRIS drone model was used and equipped with a camera, inertial sensors, Global Positioning System (GPS), and a compass. To reduce the matching time between the captured image and the georeferenced image, an adaptive window was designed. Numerous experiments were done online using all sub-models connected in a loop as in real work situations with zero cost. The RMS error of the position was less than 2.5m for both systems. Due to the adaptive window, the execution time for the correlation-based method was 40ms, which was three times less than the SIFT-based method execution time.