Abstract:

With the wide application of Unmanned Aerial Vehicle（UAV） in aerial fields such as aerial photography, surveying and mapping, environmental monitoring and express delivery, the requirements for the availability and reliability of micro-miniature drones are higher, so the function of precise positioning is essential. Con-ducting fast and robust tracking of the target is a key issue for precision positioning of the UAV. The TLD（Tracking Learning Detector） algorithm has provided an effective solution for it. Although many scholars have studied it and improved the traditional TLD algorithm, the tracking accuracy and speed of the algorithms are still difficult to meet the landing requirements of UAV. Under the framework of TLD, this paper proposes an algorithm for automatically determining the target through target’s shape features, which improves the autonomy of the positioning process. Kernelized Correlation Filter（KCF） is used as tracker in the tracking module of TLD framework. The tracker improves the real-time, accuracy and robustness of the algorithm; at the same time, a Histogram of Gradient（HOG） and Support Vector Machine（SVM） is used in the positioning process to achieve self-correction of target detection to ensure accurate target recognition. 7 types of test video simulating landing of UAV are tested to verify the performance of the proposed algorithm. Compared with the other three tracking algorithms and tested in the landing experiment, the test results show that the robustness and accuracy of the algorithm are better than other algorithms and the processing speed can reach 31.47 f/s; Therefore, the algorithm proposed uses the kernel correlation filter as the tracker in the TLD framework, combines the tracking and detection results effectively, approves the processing speed and adds self-correction of target detection to ensure the tracking accuracy, then the full autonomous precision positioning of the UAV can effectively achieve.

Key words: visual tracking, TLD（Tracking-Learning-Detection） algorithm, kernel correlation, self-correction of target detection, Support Vector Machine（SVM）

摘要：

四旋翼无人机（Unmanned Aerial Vehicle，UAV）在航拍、测绘、环境监测、快递等航空领域的广泛应用，对四旋翼无人机的可用性和可靠性提出了更高的要求，而其实现自主精准降落的功能是必不可少的。对目标进行快速鲁棒性跟踪是实现降落的重要基础，TLD（Tracking Learning Detector）算法为这一问题提供了一种有效的解决办法，虽然许多学者对其进行了研究并对传统的TLD算法进行了改进，但算法的跟踪精度及速度仍然难以满足无人机的降落要求。提出了一种基于TLD框架的目标跟踪算法来实现无人机与特定降落目标之间的相对定位。该算法在TLD框架下，提出一种基于目标形状特征自主确定降落目标的算法，提高了降落流程的自主性；用核相关滤波器（Kernelized Correlation Filter，KCF） 实现了TLD框架中的跟踪器，提高了算法的实时性、精准度及鲁棒性；同时在降落过程中采用一种基于方向梯度直方图特征（Histogram of Gradient，HOG）和支持向量机（Support Vector Machine，SVM） 的目标识别方法，以实现目标检测自矫正，保证长时间准确跟踪目标。在七类模拟无人机进行降落的视频集下验证了该算法，与其他三种跟踪算法进行对比，并进行实际降落测试。测试结果表明，该算法的鲁棒性和精准度均优于其他算法，处理速度可达到31.47?f/s，故而在TLD框架下采用核相关滤波器作为跟踪器，对跟踪及检测结果进行有效融合并提高算法实时性的同时，增加的检测自矫正环节保证了长时间跟踪的准确度，从而有效地实现了无人机全自主精准降落。

关键词: 视觉目标跟踪, TLD（Tracking-Learning-Detector）算法, 核相关滤波, 目标检测自矫正, 支持向量机（SVM）