In view of the navigation and positioning problem of Unmanned Aerial Vehicle（UAV） without GPS in rooms, an improved ORB feature optical flow algorithm is proposed, combined with ORB feature and LK pyramid optical flow. Firstly, the ORB algorithm is used to extract the feature points of each frame, and puts them into the pyramid to estimate the coordinates of these points in next frame. Secondly, the rough matching, a forward and backward tracking strategy, is carried out to filter these points. Finally, the fine matching, which is composed of the FLANN-KNN matching rule and two-way double tracking strategy, is used to filter out mismatched point sets. The algorithm performance is verified and analyzed from real-time and accuracy through the experiments, which include a variety of scene extraction effects and practical application of UAVs. The results of simulations show that the proposed improved algorithm has better positioning effect and better real-time performance.