四旋翼无人机多地面相似目标跟踪与轨迹预测

发布时间：2019-06-13 08:08

【摘要】：实时目标跟踪与轨迹预测在安防监控、智能交通、军事打击等领域有着广阔的应用前景。本课题以国际空中机器人大赛(IARC)第7代任务为背景,以四旋翼无人机为平台对该问题进行了研究,设计了一种检测器与跟踪器相融合的多相似目标跟踪算法,并以此为基础实现了目标的轨迹预测。首先,为了实现初始跟踪与跟踪失败后目标跟踪区域的自动获取,以方向梯度直方图(HOG)为特征,设计了一个基于支持向量机(SVM)的多目标检测器,并通过自举法与交叉验证法对初始的检测器进行性能改善,使得该检测器的准确率达98%以上,并且对目标的尺度、旋转变化以及环境的光照变化都有较好的鲁棒性。其次,设计了一个基于改进的Mean Shift算法的实时目标跟踪器。针对该算法需要手动选取目标模型,引入目标模型模板,实现目标模型的自动获取;针对目标跟踪状态的不明确性,制定了跟踪是否失败的多重判断条件,实现了对跟踪器状态的准确监测。再次,将上述检测器与跟踪器通过融合器相结合,设计了多相似目标跟踪算法。系统启动时逐帧检测目标,搜索潜在目标,并通过颜色分割鉴定出需要跟踪的目标,引入相应的模板,开启跟踪器;在跟踪过程中,逐帧判断目标跟踪状态,并隔帧进行目标检测,搜寻可能出现的目标并进行跟踪,获得了系统稳定与实时的性能平衡。最后,提出了一种基于Kalman滤波器与最小二乘法的目标轨迹预测算法。首先对地面移动目标进行了建模分析;然后通过摄像机标定与逆透视映射,将图像坐标系中的目标位置映射到世界坐标系中,实现目标的全局定位;最后通过Kalman滤波器与最小二乘法预测并拟合目标轨迹,为四旋翼无人机搜寻并跟踪目标提供决策依据。
[Abstract]:Real-time target tracking and trajectory prediction have broad application prospects in security monitoring, intelligent transportation, military strike and other fields. In this paper, based on the 7th generation mission of (IARC), the problem is studied on the platform of four-rotor UAV, and a multi-similar target tracking algorithm combining detector and tracker is designed, on the basis of which the trajectory prediction of the target is realized. Firstly, in order to realize the automatic acquisition of target tracking area after initial tracking and tracking failure, a multi-target detector based on support vector machine (SVM) is designed, which is characterized by direction gradient histogram (HOG). The performance of the initial detector is improved by bootstrap method and cross-verification method, so that the accuracy of the detector is more than 98%, and the target scale is also improved. The rotation change and the light change of the environment have good robustness. Secondly, a real-time target tracker based on improved Mean Shift algorithm is designed. In view of the fact that the algorithm needs to select the target model manually, introduce the target model template to realize the automatic acquisition of the target model, aiming at the uncertainty of the target tracking state, the multiple judgment conditions of tracking failure are established, and the accurate monitoring of the tracker state is realized. Thirdly, a multi-similar target tracking algorithm is designed by combining the detector with the tracker through the fusion cage. When the system starts, the target is detected frame by frame, the potential target is searched, and the target that needs to be traced is identified by color segmentation, the corresponding template is introduced, and the tracker is turned on. In the tracking process, the target tracking state is judged frame by frame, and the target detection is carried out at different frames, and the stability and real-time performance balance of the system is obtained. Finally, a target trajectory prediction algorithm based on Kalman filter and least square method is proposed. Firstly, the ground moving target is modeled and analyzed, then the target position in the image coordinate system is mapped to the world coordinate system through camera calibration and inverse perspective mapping, and the global positioning of the target is realized. Finally, the target trajectory is predicted and fitted by Kalman filter and least square method, which provides the decision basis for the four-rotor UAV to search for and track the target.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：V279

【相似文献】