Landmark extraction and state estimation for UAV operation in forest

In this paper, we present the essential problems and solutions about feature extraction and state estimation for UAV operation in GPS-denied environment - forest. Tree trunks, which are selected to be the features existing in the operational environment, help the UAV localize itself in an unknown forest. The UAV is equipped with an inertial measurement unit (IMU) and a scanning laser rangefinder (LRF). The raw laser data is preprocessed to remove outliers. A clustering algorithm based on the spatial discontinuity of the consecutive laser beams is applied to the preprocessed laser scan data. The clustered segments are characterized by a group of geometric descriptors, which are subjected to a series of thresholds to remove the false tree stems such as possible ground hit and bushes. To ensure fast and correct data association, the extracted features in consecutive scans are first aligned in rotation using the yaw angle measurement of IMU. Then the scan matching algorithm is applied to estimate the incremental rotation and translation of the UAV. The rotation and translation are fed into an IMU-driven Kalman filter to estimate the UAV position and velocity.