A performance analysis of several vision-based robot localization systems is presented for real-time Micro Aerial Vehicle (MAV) navigation tasks in touch-free, GPS denied, and high-accuracy environments. The systems were designed and utilized during our quadrotor visual control research, consisting of a local positioning system (LPS), a simplified monocular visual odometry (VO), and a stereo visual odometry ranging system. Measurement performance is evaluated through experiments performed using a membrane potentiometer sensor as reference. Feature-based image processing algorithms and motion detection methods are implemented to generate 3D position information from 2D image data. Performance analysis gave verified data for the applications of those systems. Final measurement resolution of 1mm was obtained for the LPS, and a high positioning accuracy was demonstrated for the VO systems.
展开▼