Research On Key Techniques Of Airborne Vision-based Navigation For Autonomous Landing Of A UAV On A Ship Deck

The shipborne Unmanned Aerial Vehicle(UAV) has many advantages such as wide combat range, high effectiveness-cost ratio, zero casualty and can execute tasks under severe conditions. A great emphasis has been placed on the development of the shipborne UAV by world military powers. Landing on a ship deck is a very important and difficult link which is the most accident-prone in the whole combat chain of the shipborne UAV. Therefore it has a great significance to work out effective navigation methods for the UAV landing on a ship deck as soon as possible.Vision-based navigation methods use vision sensors to capture images and obtain parameters required for navigation by image processing. These methods have many advantages such as convenient operation, low cost and high precision. In addition, they do not rely on satellites and the work electromagnetic frequencies of them are out of the frequency range of current electromagnetic countermeasure, making them adapt well to the battlefield environment. Therefore completing autonomous landing of the UAV on a ship deck independently or auxiliarily by vision-based navigation methods has become a development trend.Airborne vision-based navigation for autonomous landing of a UAV on a ship deck is taken as the research background of this dissertation. UAV attitude measurement based on horizon and sea-sky-line detection, ship target localization, automatic ship target segmentation, relative pose estimation between the UAV and the ship and other related technologies are studied systematically according to the far-to-near landing process of the UAV. The main achievements of this dissertation are as follows:1. A new horizon detection method and a new sea-sky-line detection method are proposed for computing attitude parameters of the UAV according to imaging characteristics of the horizon and the sea-sky-line, and a validity analysis method of detection results is presented. They are as follows: â‘  A coarse-to-fine horizon detection method based on between-class variance, this method uses a coarse-to-fine strategy to detect horizon and it has properties of strong stability, high precision and fast operation. The horizon in complicated background can be detected out effectively by this method. â‘¡ A sea-sky-line detection method based on black-white template correlation, this method makes full use of characteristics of the sea-sky background image and needs only to process pixels in several small region. The sea-sky-line can be detected out fast and stably by this method. â‘¢ The validity analysis method of horizon and sea-sky-line detection results judges the validity of detection results by computing the confidence level and it can distinguish correct and incorrect detection results effectively.2. Accurate ship target localization is an important content in the process of landing navigation of the UAV. Aim at the localization problem of the ship target on the sea, monocular motion trajectory intersection method is further studied. The solution method based on the criterion of minimum residual sum of squares of visual angle, and the observer trajectory optimization method based on crossbreeding particle swarm algorithm and Fisher information matrix are proposed. The optimized monocular motion trajectory intersection method can obtain better performance for locating ship target compared with traditional methods. In addition, monocular motion trajectory intersection method is expanded to multi-vision motion trajectory intersection method in theory. Solution method and observer trajectories optimization method are provided. The precision and adaptability of the multi-vision motion trajectory intersection method are better than that of the traditional first crossing then fitting method.3. Accurate separating the ship target from the sea background image is an important prerequisite of relative pose estimation between the UAV and the ship target. Aim at the automatic segmentation problem of the ship target on the sea, two automatic segmentation methods which are based on visual saliency and graph theory are proposed. They are automatic ship target segmentation method based on Context-aware and graph cut, and automatic ship target segmentation method based on Frequency-tuned and Random Walk. These two methods combine visual saliency detection methods and interactive image segmentation methods based on graph theory organically and realize the automation of human interactive segmentation methods. Both the two methods can separate the ship taret from the sea background automatically and effectively, and they have obvious advantages compared with traditional methods.4. Aim at the problem of relative pose estimation between the UAV and the ship target, two relative pose estimation methods are proposed. They are as follows: â‘  A relative pose estimation method based on the contour model of the target, this method obtains relative pose parameters between the UAV and the ship target by matching and registering the target contour in current image with target contours in model database, and it has advandages of strong stability and high precision. â‘¡ A relative pose estimation method based on infrared lamps, this method calculates the relative pose parameters between the UAV and the ship target by detecting infrared lamps which are placed on the ship target beforehand. It is convenient to use and has high precision, and thus it is suitable for engineering applications.