针对自主空中加油对接阶段锥套跟踪问题, 提出了一种基于tracking-learning-detection (TLD)的锥套跟踪算法。该算法将加油锥套的跟踪任务分解成跟踪、学习、检测3个部分。跟踪模块在LK光流法的基础上添加跟踪失败自检测, 筛选出好的跟踪点, 跟踪加油锥套; 检测模块构建级联分类器, 对滑动窗遍历得到的图像块进行分类并返回含有目标的图像块, 融合跟踪模块的跟踪框, 给出最终跟踪结果; 学习模块引入P-N约束修正错误样本并学习更新检测模块。利用Creator/Vega Prime软件对空中加油进行视景仿真, 在视景仿真视频上测试锥套跟踪算法。结果表明: TLD算法跟踪加油锥套成功率达95.5%, 处理每帧平均耗时31.4 ms, 能够满足加油锥套跟踪鲁棒性、准确率、实时性的要求。

To solve the drogue tracking problem during the capture phase of autonomous aerial refueling, a tracking-learning-detection (TLD) algorithm was proposed. The algorithm decomposes the drogue tracking task into 3 sub-tasks: tracking, learning and detection. The tracking component is based on the Lucas-Kanade (LK) method extended with failure detection and chooses tracking points with good performance to track the refueling drogue; and the detector constructs cascaded classifier to sort the picture patches scanned by scanning-window and returns patches containing object, which integrates with tracking component result and gives final tracking result; and then the P-N constraints are introduced into the learning component to correct wrong samples and then learn and update the detector. A scene simulation for aerial refueling was developed based on Creator/Vega Prime. The experimental results of the test on the video of scene simulation show that the success rate of tracking drogue is 95.5%, the average time consumption is about 31.4 ms, and the algorithm can meet well requirements of drogue tracking, such as robustness, precision and real time performance.