【作者】 袁斌；

【导师】 郝应光；

【作者基本信息】 大连理工大学 ， 通信与信息系统， 2010， 硕士

【摘要】 自主着降是无人直升机实现自主飞行的重要基础功能,基于视觉系统辅助实现自主着降是当前国内外该领域的研究主流。针对已知着降标志,提出一种识别和估计无人直升机自身状态信息的实时算法。使用经过校准的单目摄像机,通过提取相继帧中的特征点来估计瞬时的相对姿态和位置参数。该算法主要由图像预处理,地标识别,特征提取,目标跟踪和运动状态估计等组成。跟踪算法采用的是基于Lucas-Kanade的光流跟踪方法,对于相对姿态的估计,辅以惯性测量信息,较好地实现了计算结果的正确性。在实验室环境下,构建了硬件实验平台,实现了算法程序。实验结果表明,算法能满足实时性要求,其计算结果可靠,误差在允许的范围内。针对无人直升机在特殊条件下和紧急情况下的着降要求,研究了在没有先验知识的条件下,基于双目视觉技术实现在未知区域自主着降的方法。通过利用图像信息恢复三维场景中的结构信息,并经由计算出的三维场景中的平面信息,以及根据无人机着降区域的要求：平坦、斜度、面积、地表特征等因素选择安全着降区域。论述了双目立体视觉系统的标定,以及从失真校正,水平极线校正,立体匹配,三维场景结构恢复等各阶段采用算法的考量,以达到快速的处理速度。在经过反复确认的安全着降区域范围内,选择对于视角转换和光强变化等不变的特征量,建立从摄像机坐标系到固定的地面世界坐标系的转换,通过特征跟踪来确定无人机相对于安全着降区域的位置和姿态,引导无人机成功地实现在未知环境下的自主降落过程。更多还原

【Abstract】 Autonomous landing is a key capability in the autonomy of UAV. It is a major direction of development to accurately achieve self-landed based on the vision system.The paper presents a real-time algorithm of identifying landing pad and estimating the state information for landing an unmanned aerial helicopter on a given landmark automatically. The algorithm estimates the instantaneous attitude and position parameters of the helicopter relative to the landing pad from continuously tracking over feature points in consecutive frames using a calibrated monocular camera. The design of this vision system mainly performs image processing, landmark recognition, feature extraction, target tracking and motion estimation. Tracking algorithm is based on the Lucas-Kanade optical flow tracking method, and the orientation, provided by an inertial measurement unit, refines the computed attitude. The proposed algorithms are simulated on a model platform indoors in the laboratory conditions for the preliminary checking. Monocular experimental demonstrate that the algorithms can achieve real-time operation, and that the results are reliable within the range of error tolerance.Given the circumstances of emergency, a method is proposed to select a safe landing area without prior knowledge of the environments. The structure of three-dimensional scene information is reconstructed using the image information retrieval based on binocular vision system. And the scene of the plane information is calculated, combining with the requirements of landing:flat, slope, area, and surface characteristics. The calibration of stereo vision system, the distortion correction, the level of polar correction, stereo matching, and three-dimensional scene structure recovery algorithm are discussed in order to achieve the fast processing speed. After repeatedly confirmed the safety of the decided region, select the invariants to establish the fixed ground coordinate system from the camera coordinate system. It can aid the UAV successfully landing in an unknown environment through feature tracking to determine the position and orientation of UAV relative to the security zone.更多还原