Obstacle Detection for Low Flying UAS Using Monocular Camera

This paper describes an obstacle detection algorithm for low flying unmanned aircraft system (UAS) using an inertial aided inverse depth Extended Kalman Filter (EKF) framework. The EKF framework fuses inertial measurements with monocular image sensor measurements to estimate the positions of a number of landmarks as well as the position and orientation of the UAS. A high resolution sparse terrain elevation map and UAS trajectory can then be computed from the filter state vector. An inverse depth parameterization is used to describe the position of the landmarks so that features at all ranges can be tracked by the filter. A test flight was conducted to test the algorithm in a realistic scenario. The result shows that the algorithm produces accurate terrain elevation model, and is capable of generating accurate high resolution terrain elevation map when image sensor with high resolution and dynamic range is used.