响应置信度的多特征融合核相关滤波跟踪算法
2022,30(5):191-196
摘要：摘要：针对传统核相关滤波（KCF）在跟踪彩色视频序列不能有效利用颜色特征，并且处理目标遮挡和形变能力低等问题，提出一种响应置信度的多特征融合核相关滤波跟踪算法。该算法首先提取目标图像的方向直方图特征和颜色直方图特征，通过计算高响应值点在响应图上层的占比，来判断目标的跟踪情况，进而调整学习率的大小；然后用两种特征的平均峰相关能量(APCE)和最大响应峰值的乘积来加权融合目标位置。实验对比表明，提出的跟踪算法在精度和成功率上相对于KCF算法分别提升了12.8%和22.6%, 在目标发生遮挡、快速移动、旋转等复杂情况下仍然具有较强的鲁棒性。
关键词：目标跟踪；核相关滤波；响应置信度；多特征融合；鲁棒性；目标遮挡
Multi-feature Fusion Kernel Correlation Filtering Tracking Algorithm Based on Response Confidence
Abstract：Abstract：To address the problems that traditional kernel correlation filtering (KCF) cannot effectively use color features in tracking color video sequences and has low ability to deal with target occlusion and deformation, a multi-feature fusion kernel correlation filtering tracking algorithm with response confidence was proposed. The algorithm first extracts the orientation and color histogram features of the target image, to determine the tracking of the target by calculating the percentage of high response value points in the upper layer of the response map, and then adjusts the size of the learning rate; then the product of the average peak correlation energy (APCE) and the maximum response peak of the two features is used to weigh the fusion target positions. The experimental comparison shows that the tracking algorithm proposed improves 12.8% and 22.6% respectively in accuracy and success rate compared with the KCF algorithm, and still has strong robustness under complex situations such as occlusion, fast motion and rotation of the target.
Key words：object tracking; kernel correlation filter; response confidence; multi-feature fusion; robustness; object occlusion
收稿日期：2021-11-16
基金项目：国家自然科学基金项目（面上项目，重点项目，重大项目）；江西省研究生创新专项资金项目
