Abstract:

UAV object tracking is a considerable issue of computer vision. The response after multi-feature fusion usually contains noise. To solve this problem, a sub-peak suppression UAV object tracking algorithm based on multi-feature fusion is proposed. Firstly, HOG features and two-dimension color names are extracted and a response by fusion is generated. Secondly, the sub-peak response is suppressed to remove the noise, the sub-peak is aggregated into a central main peak. Finally, the adaptive model updating strategy is recommended, the robustness of the algorithm is further enhanced. Experiments are carried out on UAV123 and VisDrone2019 datasets. The results show that the algorithm has better precision and robustness in the challenging scenarios such as fast motion and viewpoint change of UAV.

Key words: computer vision, object tracking, background-awareness, sub-peak suppression, handcrafted feature

摘要：

无人机目标跟踪是计算机视觉一个热门的研究方向。多特征融合后的响应通常含有噪声，为了解决这个问题，提出一种基于多特征融合的次峰响应抑制的无人机目标跟踪算法。提取HOG特征和二维颜色属性特征，并进行融合产生响应。对次峰响应进行抑制以去除噪声，将多个次峰聚合为一个中心主峰。引入自适应模型更新策略进一步增强算法的鲁棒性。在UAV123和VisDrone2019数据集上进行实验，结果显示该算法在无人机的快速运动、视角变化等挑战场景中表现出较好的跟踪精度和鲁棒性。

关键词: 计算机视觉, 目标跟踪, 背景感知, 次峰抑制, 手工特征