【作者】 王瑶；

【导师】 高明；

【作者基本信息】 西安工业大学 ， 测试计量技术及仪器， 2012， 硕士

【摘要】 无人机较载人飞机而言具有很多优点,如制造成本低、重量较轻、体积较小、操作简单、使用方便、战场生存能力较强、隐身性能好、对作战环境要求低,并且可以准确、高效和灵便的执行各种军事或民用任务,因此备受世界各国青睐。然而,在无人机实际应用之前,必须首先对其进行靶场测试以确保无人机的性能稳定,但因其无人驾驶,需要在地而对其完成准确跟踪,才能方便地面人员了解无人机的状态。本文设计了一套基于合作目标的无人机自动跟踪算法,该算法能够对图像进行预处理,对图像序列进行目标检测与目标识别,根据闽值分割后的图像提取目标有用信息,利用目标的运动模型完成目标轨迹的预测,并在预测位置附近搜索目标。主要做了以下工作：分析和比较了无人机的跟踪方案,采用合作目标配合无人机的跟踪；通过对两种经典去噪方法进行对比分析,采用改进的中值滤波方法去除了图像中的噪声,对去噪后的灰度图像进行数学形态学处理,有效调整了图像中的明暗细节,使图像更加平滑；介绍了目标检测的方法,并根据分层投票表决法判断日标真伪,对含有真实目标的图像进行阈值分割。采用基于迭代的改进最大类间方差(Otsu)法,较完整地提取出了图像中的光斑,然后对光斑进行定位；利川Kalman滤波算法对目标轨迹进行预测,在预测位置附近,根据直方图信息进行K-S检验,判断当前搜索区域是否与目标模板匹配,并且给出了仿真结果；介绍了跟踪过程中所需要的坐标系及其之间的变换关系,并给出了激光器及CCD摄像机姿态的调整原则。实验及仿真结果表明：基于选择排序法的中值滤波可以更加快速的去除图像中的噪声,改进的Otsu法可以解决图像分割时造成的误分割问题;识别目标后,Kalman滤波算法能够对目标轨迹进行有效预测,基于K-S检验的直方图匹配方法可以在预测位置附近找到最佳匹配位置,从而实现目标跟踪。这种目标跟踪算法能够满足目标跟踪的要求,不但可以有效识别目标,还能够提高目标跟踪的效率。更多还原

【Abstract】 Compared with manned aircraft, UAV has many advantages, such as low cost, light weight, small volume, simplicity of operation, usableness, strong survivability, good stealth, low requirements for battle environment, and it can perform various military or civilian tasks accurately, efficiently and agilely. UAV is strongly favored by the countries all over the world. However, UAV must be range tested to ensure performance stable before actual application. Because of unmanned driving, UAV needs to be tracked accurately by researchers on the ground, which can help people to understand the state of the UAV.TXX UAV automatic tracking algorithm based on cooperative object is designed in this thesis. What can be realized by this algorithm are image preprocessing, target detection and recognition of image sequence, extraction of target useful information on the basis of the image threshold segmented, prediction of target track by using moving model of the target, and target searching near the prediction location.The tracking schemes of UAV are analyzed and compared, with the cooperative object being used to cooperate to track UAV. By the contrast and analysis of two classical de-noising methods, the improved median filter method is used to remove the noise off the image. Mathematical morphology processing is applied to adjust the brightness details to make de-noising image smoother. Target detection methods are introduced, and layering voting is used to identify the verity of target. The image containing true target is threshold segmented through improved Otsu method based on iteration. The facula can be extracted from the image completely and then be located. Target track is predicted by Kalman filtering algorithm. Near the prediction position, whether the current search area matches with the target template can be judged by using the K-S test based on histogram information, and simulation results are given. Finally, coordinate system and their transformation relations in tracking process are presented, and the attitude adjustment principle of lasers and CCD camera is given.The experiment and simulation results indicated that the noise was removed off the image faster through median filter based on selection sort, and the problem that the target and background were segmented mistakenly was solved by improved Otsu method. After target recognition, the target track could be predicted effectively by using Kalman filtering algorithm. The optimal matched position was found near the prediction position through the histogram matching method based on K-S test. Finally, target tracking could be realized with this tracking algorithm. The requirement of target tracking was satisfied by this target tracking algorithm. The target could be recognized effectively, and also the efficiency of target tracking was improved.更多还原