基于注意力机制和特征匹配的空对地多目标检测与跟踪算法

多目标跟踪算法是实现无人机自主导航的关键技术，为解决现有方法存在的小目标检测能力弱、计算能耗大、鲁棒性差等问题，提出一种基于注意力机制和特征匹配的多目标空对地跟踪算法，以实现航拍视角下对目标的精准高效跟踪。首先，引入通道可分离卷积，实现目标检测模型的轻量化；其次，构造融合空间注意力机制的小目标检测分支，提高对小微目标的检测精度，最后，优化目标跟踪算法的外观重识别网络，提高多目标跟踪效率。使用Visdrone2019-MOT数据集对所提算法进行验证，实验结果表明，所提算法的MOTA值提高了0.６%，FPS值为21.31帧/s，在模型大小和跟踪精度上实现了较好的平衡。

Air-to-ground multi-object detection and tracking algorithm based on attention mechanism and feature matching

Multi-object tracking algorithm is an essential technology for achieving autonomous navigation of unmanned aerial vehicles. In order to solve the problems of weak detection ability for small target, high computational energy consumption, and poor robustness in existing methods, we propose a multi-object air-to-ground tracking algorithm based on attention mechanism and feature matching, which aims to achieve accurate and efficient tracking of targets from an aerial perspective. Firstly, we introduce channel separable convolution to achieve lightweight object detection models. Secondly, we construct a small target detection branch which integrates spatial attention mechanism in order to improve the detection accuracy of small and micro targets. Finally, we optimize the appearance recognition network of the target tracking algorithm to improve the efficiency of multi-object tracking. The proposed algorithm is validated using the Visdrone2019-MOT dataset,according to the results,the proposed algorithm achieves a MOTA value of 43.1% and an FPS value of 21.31 frames/s, which indicates a good balance between model size and tracking accuracy.