摘要
在进行跟踪任务时,当运动目标发生形变或旋转以及受到光照变化或背景干扰时,会发生偏移现象或丢失跟踪目标,从而导致跟踪精度降低。据此,提出基于多特征融合和改进SIFT的目标跟踪算法。在图像的高熵部分进行特征点的提取,并使用哈希算法将错误匹配的特征点剔除。同时对感知哈希和差异哈希进行改进,将改进后的图像哈希特征、颜色特征和SIFT特征进行融合并应用于跟踪算法。将算法在OTB-100数据集上进行实验,成功率达到了94.3%。
When the moving target was deformed,rotated and interfered by the illumination or background during the tracking task,it could be shifted or lost,which would reduce the tracking accuracy.A target tracking algorithm based on multiple feature fusion and improved SIFT was proposed.The feature points were extracted in the high entropy part of the image,and the mismatched feature points were eliminated by using the hash algorithm.At the same time,the perceptual hash and differential hash were improved,and the improved image hash features,color features and SIFT features were fused and applied to the tracking algorithm.The algorithm was tested on OTB-100 dataset,and the success rate reacheed 94.3%.
作者
李文举
王子杰
崔柳
LI Wenju;WANG Zijie;CUI Liu(College of Computer Science and Information Engineering,Shanghai Institute of Technology,Shanghai 201418,China)
出处
《郑州大学学报(理学版)》
CAS
北大核心
2024年第1期40-46,共7页
Journal of Zhengzhou University:Natural Science Edition
基金
国家自然科学基金项目(61973307,61903256)。
关键词
目标跟踪
图像哈希
信息熵
颜色矩
SIFT
target tracking
image hash
information entropy
color moment
SIFT