Research On Vision-based Landing Technology For An Unmanned Aerial Vehicle

Due to the high probability of accidents, the demand of precision, reliability and real-time of landing guidance is crucial in the flight, as the usage of UAV gets wider in the civil and military field, the automatic landing guidance is becoming an urgent problem to be solved.The vision-based guiding system which takes advantage of the traditional photogrammetry, optical measurement, modern computer vision and digital image processing, has the quality of high precision, non-contact, anti-jamming and low cost, showing good application perspective in the UAV's landing. High precision calibration should be executed beforehand in the vision guidance ,then tracks and matches the targets during the landing process in real time, calculates the position and attitude by multi-station intersection or pose estimation algorithm, feeds the information back to the control loop and guides the UAV to the runway smoothly.The article focuses on the key technology in the landing of UAV, particularly the wide-range camera calibration, the geometric constraints of multi-camera relay measurement and its application in targets tracking and matching. The results have been successfully applied in the landing system for UAV developed by the author's department to compete in the AVIC cup-International UAV Innovation Grand Prix.The main work in this article is:1. Take a survey of the guiding development of the airplane in the civil and abroad, analyze and compare the strongpoint and drawbacks, combining the vision technology, proposed the vision-based assistant landing method for the UAV.2. The camera projection model is analyzed and the invariants in the geometry transformation are listed.3. To cope with the problem of wide-range camera calibration, two efficient methods based on simplified camera model and infinite homography respectively have been proposed in this paper, and the pattern-based calibration method is introduced to solve the calibration of intrinsic camera parameters. Simulation and experiments demonstrate the efficient of the methods above.4. Multi-camera relay measurement is a good method to solve the problem of reduction of precision due to the increase of the distance, the geometric constraints between the multi-view images are studied, and trifocal tensor which reveals the relationships of three views is introduced to solve the targets prediction and matching , also simulation proves the efficiency.5. Based on the algorithms studied above, a vision-based assistant landing system is proposed, the modules of the system are developed, the hardware is selected. Experiments shows that the system well satisfies the landing precision, and has good application perspective.