Position Estimation of Small UAV from Monocular Camera by Using Matched Features

This paper focuses on position estimation of a small unmanned aerial vehicle (UAV) using a monocular camera. Features from accelerated segment test (FAST) descriptors are used as a matched pattern to estimate differential change in position of the UAV. Visual simultaneous localization and mapping (V-SLAM) is a probabilistic filter-based method and a prominent real-time positioning method in robotics. V-SLAM performs drift-free tracking of the pose of a UAV on long run but the prediction states have limited certainty because of using sparse number of features used in real-time position estimation. Visual odometry (VO) is a deterministic positioning method and is more accurate to estimate the relative UAV position from adjacent frames without a persistent map. VO gives high drift error on the long run because of the accumulation of drift error at each frame transformation. Bundle adjustment (BA) and loop closure are two error optimization techniques to reduce drift error in VO. Due to limited computation resources available in the small-scale UAV, the optimization techniques are not appropriate in the small UAV positioning. In this work, VO with fractional V-SLAM is proposed to reduce the drift error on position estimation from matched features. The obtained results show the positioning estimation from proposed method works in an outdoor environment and that overperforms than VO and V-SLAM.