1-Point RANSAC for extended Kalman filtering: Application to real-time structure from motion and visual odometry

This paper presents a framework for integrating sensor information from an inertial measuring unit (IMU), global positioning system (GPS) receiver, and monocular vision camera mounted to a low-flying unmanned aerial vehicle (UAV) for producing large-scale terrain reconstructions and classifying different species of vegetation within the environment. The reconstruction phase integrates all of the sensor information using a statistically optimal nonlinear least-squares bundle adjustment algorithm to estimate vehicle poses simultaneously to a highly detailed point feature map of the terrain. The classification phase uses feature descriptors based on the color and texture properties of vegetation observed in the vision data and uses the terrain information to build a georeferenced map of different types of vegetation. The resulting system can be used for a range of environmental monitoring missions such as invasive plant detection and biomass mapping. Experimental results of the algorithms are demonstrated in a “weed-finding” mission over a large farmland area of the Australian outback. © 2010 Wiley Periodicals, Inc.