Cameras are used for the navigation system in lots of vehicles such as land vehicles, small-sized and low-cost airborne systems and mobile robots lately. Even though camera-based vision navigation system is sensitive to light condition and suffers from high computational burden, it has greater advantages like light weight, low-cost and low-power consumption.
This paper studies two main subjects related to the landmark-based vision navigation system. Firstly, the geometric effects has been investigated for the landmark geometry in the two-dimensional and three-dimensional vision navigation system. The relationship between the landmark measurement error on the camera focal plane and the navigation error is derived for landmark geometry by perturbation approach. And then DOP(Dilution of Precision) has been expressed separately into the position DOP and the attitude DOP. Values of each DOP have been observed for the various configurations of landmarks. The results show that the position DOP increases as the landmark distance from the vehicle increases and the position DOP and the attitude DOP decrease when the angles between landmarks from vehicle becomes bigger. It is also observed the effects on navigation accuracy for the camera focal length. These DOP research shall be very useful to place artificial landmarks or to select landmark combination to improve navigation performance in the vision navigation and its integrated navigation system.
Secondly, it is proposed the INS(Inertial Navigation System) and vision integrated navigation system considering the vision navigation error for the landmark geometry whose effect is studied in the paper. The INS/vision integration system can make its synergistic relationship. The proposed system is designed to use the landmark measurements to have smallest DOP values in order to improve navigation accuracy. In addition, the system uses focal plane measurements of landmarks for effective usages in an area which has fewer visible landmarks. The integration model is designed to use both sensors complementally in the system. The navigation performance of the proposed system is evaluated through Monte-Carlo simulations and ground-vehicle experiments comparing the other integration model which uses the vision navigation solution. It is shown from the results that the proposed design can give better performance and show robustness in weak landmark environments in particular.
Further researches shall be vision-based context adaptive navigation and a more deeper INS/vision integration method.