基于混合特征融合的高效孪生单目标跟踪方法

打开文本图片集

关键词：单目标跟踪；孪生网络；特征融合；特征细化；特征聚合 中图分类号：TP391.4 文献标识码：A doi：10.37188/CJLCD.2025-0097 CSTR：32172.14.CJLCD.2025-0097

Abstract：To balance the trade-ofbetween tracking accuracy and model complexity，an eficient singleobject tracking method is proposed based on a siamese network. The method employs a lightweight MobileNet-V3 as the backbone network，significantly reducing the computational load and number of parameters for feature extraction. Additionally，a hybrid feature fusion module is designed，comprising a rapid feature refinement unit and a dual-branch feature aggregation unit. The rapid feature refinement unit effectively decreases the number of feature vectors by aggregating queries and optimizing keys，thereby quicklyextracting key information about the target object. Thedual-branch feature aggregation unit further enhances tracking performance through a multi-head attention mechanism that fuses features from different branches.Comparative experiments with other tracking algorithms are conducted on the LaSOT， OTB10O，and UAV123 datasets. Experimental results demonstrate that the proposed method maintains satisfactory tracking performance while exhibiting lower model complexity. Furthermore，it sustains robust tracking capabilities in various complex scenarios，including fast motion and rotation.

Key words： single object tracking； Siamese network； feature fusion; feature refinement；feature aggregation

1引言

随着计算机视觉的快速发展，目标跟踪已成为该领域的重要任务之一。（剩余17565字）