基于多特征融合的红外目标跟踪方法研究

Research on Infrared Target Tracking Method Based on Multi-feature Fusion

摘要

针对复杂场景下因红外图像成像质量差、纹理模糊、易受亮目标干扰等而导致红外目标跟踪效果不佳的问题，提出一种基于多特征自适应融合的红外目标跟踪方法。设计一种大核小波网络（large kernel & wavelet network，LKWNet）用于提取深度特征，设计自适应方向梯度直方图算法用于提取手工特征，并将两种特征相结合，提高红外图像特征表达能力。实验结果表明，基于本文特征提取方法设计的红外目标跟踪器 ECOIR 在 LSOTB-TIR 数据集上的 Succ.AUC和Prec@20分别为 79.0% 和 65.0% ，在 PTB-TIR 数据集上的 Succ.AUC 和 Prec@20 分别为 81.9% 和 63.2% ，表现出较好的鲁棒性和跟踪精度。

Abstract

To solve the problem of poor infrared target tracking performance caused by poor imaging quality，fuzzy texture and susceptibility to interference from bright targets in complex scenes，an infrared target tracking method based on adaptive fusion of multiple features is proposed.A large kernel & wavelet network（LKWNet）is designed to extract deep features，and an adaptive histogram of oriented gradients method is designed to extract hand-crafted features.The two features are combined to improve the feature expression ability of infrared images.The experimental results show that the Succ.AUC and Prec@20 of the infrared target tracker ECOIR designed based on the proposed features extracting method are 79% and 65.0% on the LSOTB-TIR dataset，and 81.9% and 63.2% on the PTB-TIR dataset，respectively.It shows good robustness and tracking accuracy.

关键词

红外目标跟踪 / 自适应融合 / 深度特征 / 方向梯度直方图

Key words

infrared target tracking / adaptive fusion / deep features / histogram of oriented gradients

参考文献

[1] Yuan D, Zhang H P, Shu X, et al. Thermal infrared target tracking:a comprehensive review[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 73:5000419.

[2] 李清忠. 基于两阶段时空加权特征的红外目标跟踪算法[J]. 红外技术, 2025, 47(4):437-444.

[3] Li Q Z. Infrared object tracking algorithm based on two-stage spatiotemporal weighted features[J]. Infrared Technology, 2025, 47(4):437-444. (in Chinese)

[4] Zhang W X, Huang B R, Meng S J, et al. Infrared fast-maneu- vering target tracking based on robust exact differentiator with improved particle filter[J]. Journal of Robotics, 2022, 2022 (1):6406672.

[5] Wu F Y, Liu A R, Zhang T F, et al. Saliency at the helm: steering infrared small target detection with learnable kernels[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 63:5000514.

[6] Wang H, Hu Y, Wang Y, et al. Infrared small target detection based on weighted improved double local contrast measure[J]. Remote Sensing, 2024, 16(21):4030.

[7] Fan J, Wei J B, Huang H, et al. IRSDT:a framework for infrared small target tracking with enhanced detection[J]. Sen- sors, 2023, 23(9):4240.

[8] Ding M, Chen W H, Cao Y F. Thermal infrared single-pedestrian tracking for advanced driver assistance system[J]. IEEE Transactions on Intelligent Vehicles, 2023, 8(1):814-824.

[9] Zhang H, Yin Z Y, Zhang H L. Thermal infrared object tracking using correlation filters improved by level set[J]. Signal, Image and Video Processing, 2023, 17(3):791-797.

[10] 孙梦宇, 王鹏, 徐均琪, 等. 自适应信息选择的变尺度相关滤波红外目标跟踪算法[J]. 光子学报, 2023, 52(12): 226-238.

[11] Sun M Y, Wang P, Xu J Q, et al. Adaptive information selection for infrared object tracking with variable scale correlation filter[J]. Acta Photonica Sinica, 2023, 52(12):226-238. (in Chinese)

[12] 刘刚阳, 胡博, 王宇庆. 基于改进 ECO-HC 的目标跟踪方法[J]. 液晶与显示, 2023, 38(8):1118-1127.

[13] Liu G Y, Hu B, Wang Y Q. Target tracking method based on improved ECO-HC[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(8):1118-1127. (in Chinese)

[14] 王宁, 吴伟, 王元元, 等. 多特征融合的无人艇视觉小目标鲁棒跟踪[J]. 中国舰船研究, 2024, 19(5):65-78.

[15] Wang N, Wu W, Wang Y Y, et al. Multi-feature fusion-based robust tracking of small targets in unmanned surface vehicle vision[J]. Chinese Journal of Ship Research, 2024, 19(5): 65-78. (in Chinese)

[16] 宫华, 张众圭, 胡雨桐, 等. 基于改进 YOLOv7-DeepSort 的红外视频多目标跟踪[J]. 沈阳理工大学学报, 2024, 43 (6):20-27.

[17] Gong H, Zhang Z Y, Hu Y T, et al. Infrared video multi target tracking based on improved YOLOv7-DeepSort[J]. Journal of Shenyang Ligong University, 2024, 43(6):20-27. (in Chinese)

[18] 熊偌炎, 张上, 张岳. 多尺度特征响应融合的复杂场景红外目标跟踪[J]. 计算机工程与应用, 2025, 61(12):243-257.

[19] Xiong R Y, Zhang S, Zhang Y. Multi-scale feature response fusion for complex scene infrared target tracking[J]. Com- puter Engineering and Applications, 2025, 61(12): 243-257. (in Chinese)

[20] Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]//2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition(CVPR2005).San Diego,CA,USA:IEEE,2005:886-893.

[21] Fernandez A, Mali A. TeLU activation function for fast and stable deep learning[PP/OL]. arXiv(2024-12-28)[2025-07-10]. https://doi.org/10.48550/arXiv.2412.20269.

[22] Xu G P, Liao W T, Zhang X, et al. Haar wavelet downsampling:a simple but effective downsampling module for semantic segmentation[J]. Pattern Recognition, 2023, 143:109819.

[23] Liu Q, Li X, Yuan D, et al. LSOTB-TIR:a large-scale highdiversity thermal infrared single object tracking benchmark[J]. IEEE Transactions on Neural Networks and Learning Systems, 2024, 35(7):9844-9857.

[24] Chen J, Kao S H, He H, et al. Run,don't walk:chasing high- er FLOPS for faster neural networks[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Vancouver,BC,Canada:IEEE, 2023:12021-12031.

[25] Wang B R, Yuan J, Zhang X B, et al. EASATrack:infrared object tracking based on edge activation and scale adaptive cross-correlation[C]//2023 5th International Conference on Robotics,Intelligent Control and Artificial Intelligence(RIC-AI).Hangzhou,China:IEEE,2023:801-805.

[26] Song J F, Xu H, Ren G D, et al. Infrared object-tracking algorithm with temporal features fusion Siamese network[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 73:1-13.

基金资助

辽宁省应用基础研究计划项目(2023JH2/101300239)