Abstract: Object detection and multi-target tracking technologies are becoming increasingly mature. However, when performing aerial multi-target tracking tasks in complex scenarios, issues such as small target size, large size variation, and occlusion still lead to unsatisfactory detection and tracking performance. Therefore, this paper proposes an aerial multi-target tracking algorithm based on an improved YOLOv8 and ByteTrack (YBTrack). First, a detector (MSA-YOLO) is constructed. The original convolution in YOLOv8 is replaced with a space-depth convolution, which transforms spatial information into channel dimensions, effectively preserving target details and reducing missed and false detections caused by information loss during multi-scale feature map fusion. At the same time, a lightweight accelerated space-channel attention module is designed for neck convolution to reduce computational complexity. This module also acts as a feature refinement module before the detection head, further enhancing the ability to extract target feature information. Next, to improve tracking performance, the ByteTrack tracking model is optimized. A spatial-appearance similarity matrix (ASM) is designed to enhance the model's ability to distinguish similar targets. Additionally, a target correction function is proposed to reduce the error accumulation of the Kalman filter, decreasing target offset and loss rates. Finally, the MSA-YOLO and the optimized ByteTrack are combined for multi-target tracking experiments. MSA-YOLO achieves a 9.4% improvement in mAP_0.5 on the VisDrone2019-DET dataset. The multi-target tracking algorithm improves MOTA by 11.2% and 8.3% and IDF1 by 8.9% and 7.4% on the VisDrone2019-MOT and MOT17 datasets, respectively. Experimental results demonstrate the significant tracking performance of the proposed method. Furthermore, comparison experiments with other multi-target tracking algorithms also confirm the superiority of the proposed algorithm.

摘要： 目标检测与多目标跟踪技术日益成熟，但在复杂场景下执行航拍多目标跟踪任务时，目标尺寸小、尺寸变化大、遮挡等问题仍会导致检测与跟踪效果不理想。为此，提出一种基于改进YOLOv8与ByteTrack的航拍多目标跟踪算法（YBTrack）。首先，构建了检测器(MSA-YOLO)；设计空间-深度卷积替换YOLOv8原有卷积，将空间信息转换为通道维度，有效保留目标细节，减少了多尺度特征图融合过程中信息丢失导致的漏检、误检；同时设计轻量加速空间-通道注意力模块，用于颈部卷积，降低了计算复杂度，并作为检测头前的特征细化模块，进一步增强对目标特征信息提取的能力。然后，为提高跟踪效果，对ByteTrack跟踪模型进行优化；设计空间-外观相似度矩阵（ASM），提升了模型区分相似目标的性能；并提出目标校正函数，减少卡尔曼滤波器产生的误差积累，降低了目标偏移、丢失率。最后，将MSA-YOLO与优化后的ByteTrack结合，开展多目标跟踪实验；其中，MSA-YOLO在VisDrone2019-DET数据集上，mAP_0.5提高了9.4%；多目标跟踪算法在VisDrone2019-MOT、MOT17数据集上 MOTA分别提升了11.2%和8.3%，IDF1分别提升了8.9%和7.4%，实验结果表明本文所提方法跟踪效果显著。此外，与其他多目标跟踪算法的对比实验也证明了本文算法的优越性。