Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (14): 142-150.DOI: 10.3778/j.issn.1002-8331.2204-0136

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Object Tracking Algorithm for Multi-Scale Channel Attention and Siamese Network

WANG Shuxian, GE Haibo, LI Wenhao   

  1. School of Electronic Engineering, Xi’an University of Posts and Telecommunications, Xi’an 710121, China
  • Online:2023-07-15 Published:2023-07-15

多尺度通道注意与孪生网络的目标跟踪算法

王淑贤,葛海波,李文浩   

  1. 西安邮电大学 电子工程学院,西安 710121

Abstract: Based on the Siamese object tracking algorithm of fully convolutional classification and regression, a target tracking algorithm fused with multi-scale channel attention(SiamMCA) is proposed. Firstly, the algorithm is based on the Siamese network architecture. The backbone network ResNet50 is improved, which conbined with multi-scale channel attention for feature extraction and enhancement. Then, the deep cross-correlation network is adopted to decode and track the feature map. At last, it performs fusion, classification and regression successfully. By making full use of the semantic information integration function of multi-scale channel attention, spatial information and motion information are integrated, which improves the performance of the tracker. Finally, experiments on OTB100, LaSOT long-term benchmarks, and VOT2016 datasets show that SiamMCA achieves higher accuracy, success rate, and performance than other state-of-the-art trackers. Compared with other trackers, the success rate and accuracy are higher in complex scenes, especially in fast motion, occlusion, similarity interference and scale changes.

Key words: Siamese, multi-scale, channel attention, feature enhancement

摘要: 在全卷积分类和回归的Siamese目标跟踪算法的基础上,提出了一种融合多尺度通道注意力的目标跟踪算法(Siamese multi-scale channel attention,SiamMCA)。算法基于Siamese网络架构,以多尺度通道注意力融合改进后的ResNet50作为骨干网络进行特征提取与增强,并利用深度互相关网络对特征图进行解码和跟踪,最终成功进行融合、分类和回归。SiamMCA通过充分利用多尺度通道注意力的语义信息整合功能,整合了空间信息和运动信息,提升了跟踪器的性能。最终分别在OTB100、VOT2016数据集上和LaSOT长期基准上的实验表明,SiamMCA与其他先进的跟踪器相比取得了更高的精度、成功率和性能表现,尤其是在快速运动、遮挡、相似性干扰、尺度变化等复杂场景中。

关键词: Siamese, 多尺度, 通道注意力, 特征增强