Statistical Algorithm for Attitude Estimation from Real-Time Aerial Video

Abstract

Almost all autonomous unmanned aerial vehicles are used for reconnaissance and intelligence gathering roles. This means that cameras, video transmitters, and/or video recorders are already integrated in the system and are part of the unmanned aerial vehicle payload. Current attitude estimation sensors are expensive, heavy, and consume more power than most micro aerial vehicles can tolerate. Vision-based attitude estimation can be used to augment inertial sensors for increased accuracy, or as primary pitch and roll sensing resulting in reduced vehicle cost, size, and weight. This paper presents a fast, real-time algorithm to estimate pitch and roll angles for an aerial vehicle from video frames captured using a downward-pointingcamerawithamountedfisheyelens.Thefisheyelensisinstalledtoensure the visibility of most of the earth’s horizon at sufficient altitudes. Attitude angles are estimated by the horizontal and vertical movement of the horizon circle, which moves in relation to the center of the video frame image. The system was tested and implemented on a radio controlled plane and the results proved to be successful with over 85% of the results within ±3 ◦ when compared to a traditional inertial measurement unit.