| In this research, we deal with the problem of obstacle detection for unmanned aerial vehicles using monocular vision systems. First, a fast and robust technique for precise detection of the horizon path is proposed. The method is based on existence of unique light fields that occur in imagery where the horizon is visible. The horizon path can be extensively used for navigation purposes. Furthermore, since the horizon line divides the scene into two parts, namely sky and ground, this can be very helpful in speeding up the analysis procedures. In other words, the obstacles inside the sky and on the ground can be processed separately and with different parameters. This will increase the speed or alternatively enable a more accurate analysis given the available time frame.;Finally the two components, horizon detection and obstacle detection, are coupled to form a system with the potential of rapidly processing the sky and locating obstacle with significant accuracy. The system is tested on numerous real video sequences. The results confirm the robustness of the proposed system.;In the next part, a new method for obstacle detection using optical flow is proposed. The method employs a highly efficient and accurate adaptive motion detection algorithm for determining the regions in the image which are more likely to contain obstacles. These regions then have optical flow performed on them. We call this method targeted optical flow. To compensate for leaving out sections of the objects, obstacle reconstruction is carried out using clustering. The proposed technique is significantly faster and more accurate than using standard optical flow alone. |
