河北工业科技2025，Vol.42Issue(5)：412-420,9.DOI:10.7535/hbgykj.2025yx05002
多尺度自注意力和局部匹配的光流估计方法
Multi-scale self-attention and local matching optical flow estimation method
摘要
Abstract
To address the issues of limited receptive field and edge blurring in optical flow estimation,an optical flow estimation model based on multi-scale self-attention and local feature matching was proposed.This model was an improvement upon the recurrent all-pairs field transforms(RAFT)model.Firstly,a multi-scale self-attention mechanism was integrated into the feature extraction module,which learned the dependencies between long-distance pixels using multi-scale self-attention to obtain image feature information.Secondly,a local matching module was added during the upsampling process of low-level optical flow to generate high-resolution optical flow.Then,the model was trained on optical flow estimation datasets.Finally,ablation experiments and comparative experiments were conducted on the trained model.The results show that the proposed model achieves average end point error(AEPE)of 1.18 and 1.67 on the MPI Sintel Clean and MPI Sintel Final datasets,respectively,and 1.01 and 3.40%for average end point error and flow error of all(Fl-all)on the KITTI-2015 dataset,all outperforming RAFT.The proposed optical flow estimation model exhibits high accuracy in optical flow estimation,which can provide effective support for computer vision tasks relying on high-precision motion information.关键词
计算机神经网络/光流估计/注意力机制/上采样/局部匹配Key words
computer neural network/optical flow estimation/attention mechanism/upsampling/local matching分类
信息技术与安全科学引用本文复制引用
李春华,李晓东..多尺度自注意力和局部匹配的光流估计方法[J].河北工业科技,2025,42(5):412-420,9.基金项目
中央引导地方科技发展资金项目(246Z0109G) （246Z0109G）
