The usual approaches to unmanned air vehicle(UAV)-to-ground target geo-location impose some severe constraints to
the system, such as stationary objects, accurate geo-reference terrain database, or ground plane assumption. Micro air
vehicle(MAV) works with characteristics including low altitude flight, limited payload and onboard sensors' low
accuracy. According to these characteristics, a method is developed to determine the location of ground moving target
which imaged from the air using monocular camera equipped on MAV. This method eliminates the requirements for
terrain database (elevation maps) and altimeters that can provide MAV's and target's altitude. Instead, the proposed
method only requires MAV flight status provided by its inherent onboard navigation system which includes inertial
measurement unit(IMU) and global position system(GPS). The key is to get accurate information on the altitude of the
ground moving target. First, Optical flow method extracts background static feature points. Setting a local region around
the target in the current image, The features which are on the same plane with the target in this region are extracted, and
are retained as aided features. Then, inverse-velocity method calculates the location of these points by integrated with
aircraft status. The altitude of object, which is calculated by using position information of these aided features,
combining with aircraft status and image coordinates, geo-locate the target. Meanwhile, a framework with Bayesian
estimator is employed to eliminate noise caused by camera, IMU and GPS. Firstly, an extended Kalman filter(EKF)
provides a simultaneous localization and mapping solution for the estimation of aircraft states and aided features location
which defines the moving target local environment. Secondly, an unscented transformation(UT) method determines the
estimated mean and covariance of target location from aircraft states and aided features location, and then exports them
for the moving target Kalman filter(KF). Experimental results show that our method can instantaneously geo-locate the
moving target by operator's single click and can reach 15 meters accuracy for an MAV flying at 200 meters above the ground.
|