Autonomous stereo vision based collision avoid system for small UAV**

Sense and Avoid has been recognized as one critical prerequisite for UAV's safety integration into the national airspace. Especially for small UAVs, the complex low altitude environment may impose great threat to its safety operation. In this paper, we proposed an autonomous UAV collision avoid method based on stereo vision which is valid for low altitude scenario. Initially, the scenarios is classified into the in-the-air scenario and near- the-ground scenario based on a horizon detector. Then different obstacle detection and avoidance maneuver methods are adopted. For the near-the-ground scenario, a 3D reconstruction based detection method and a virtual obstacle plane based collision avoid method are adopted to sense and avoid the static obstacles on the ground. For the in-the-air scenario, small target detection and triangulation are used to localize the flying intruders and a collision cone based avoidance method are adopted. In the end, the stereo vision based SAA system are descirbed and hardware-in-the-loop test are carried out to validate the proposed collision avoid framework. The results shows that stereo vision can be an efficient way to assist the sense and avoid function for small UAVs operation