摘要
TLD(tracking-learning-detection)跟踪算法在目标作平面外旋转、快速移动和非刚性形变的情况下易跟踪失败,而核相关滤波器(kernelized correlation filters,KCF)跟踪算法可以有效应对上述跟踪情景但缺乏跟踪失败恢复机制,导致目标重新出现后无法继续跟踪。针对以上问题,通过有效结合这两种算法,提出一种基于TLD框架下的核相关滤波器跟踪检测算法。在跟踪模块中融入颜色特征,进一步增强算法的整体跟踪性能。通过在不同视频序列上进行对比实验,结果表明,与原算法相比,改进后的算法可以长时间准确地跟踪目标,并具有更高的成功率。
Tracking-learning-detection (TLD) tracking algorithm may fail in case of fast motion, full out-of-plane rotation and non-rigid deformation of target, while kernel correlation filter (KCF) tracking algorithm can effectively deal with the above-mentioned tracking situations, however, it lacks tracking failure recovery mechanism, so it is hard to resume tracking once the disappeared object reappears. Aiming at the above problems, an efficient combination of the two tracking algorithms was presented to devise a KCF target tracking algorithm based on TLD frame- work. Moreover, color feature was incorporated into the tracker to further boost the overall tracking performance. Finally, the results of contrast experiments on different video sequences show that the improved algorithm can track the target accurately for a long time and it has a higher success rate.
出处
《应用科技》
CAS
2018年第1期77-83,共7页
Applied Science and Technology
基金
国家自然科学基金项目(61003128)
关键词
目标跟踪
TLD
核相关滤波器
特征融合
循环矩阵
跟踪成功率
跟踪精度
长期跟踪
object tracking
TLD
kernel correlation filters
feature integration
circulant matrices
tracking success rate
tracking precision
long-term tracking