Inertial Navigation System Assisted Visual Localization

With recent advancements in Global Positioning Systems (GPS), localization systems are now typically equipped with a GPS. However, in a large variety of environments and real-world applications, GPS-based localization systems are not practical. This research tackles such a problem and illustrates the idea of fusing a camera and an inertial navigation system (INS) to create a dead reckoning localization system. The original purpose of the localization system is for a pipe inspection robot, but the proposed concepts can be readily applied to any environment where there is a wall in close proximity. The proposed sensor system can determine motions with up to six degrees of freedom using a camera and an INS. The system must assume a geometry for the wall, such as a at surface, a wall in a hallway, or the round surface of the inside of a pipe. If the geometry of the wall is unknown then another sensor, such as a laser range nder, can be added to measure the range and estimate the overall shape of the wall. The localization system uses a combination of optical ow and image registration to obtain information about six degrees of freedom from a wall with little to no features. The INS provides an estimated motion for the camera system. The estimation of the motion is used by the optical ow algorithm to reduce the computational load signi cantly and by the image registration to decrease the likelihood of the algorithm diverging. The system is validated using numerical simulation and experiments. The experiments were conducted on a test platform constructed speci cally in this research project to simulate the motion of a free-swimming robot inside a pipe. The simulator uses accurate ii position sensors to measure the exact location of the proposed localization system and compare it with the results obtained from the latter. Both the numerical simulation results and the results from the simulator are in agreement with reading of the localization system developed in this research.