联合检测的自适应融合目标跟踪

Adaptive fusion target tracking based on joint detection

由于各种因素的干扰,在现实复杂的情况下目标跟踪过程中可能出现模型漂移和跟踪失败等问题,针对目标跟踪的鲁棒性和准确性提出一种联合检测的自适应融合目标跟踪算法。根据深层和浅层卷积特征具有的不同优点,使它们单独作用于相关滤波器得到其各自的响应分数,通过最小化损失使不同卷积特征的响应分数自适应融合。结合本研究的位置检测方法判断预测位置的有效性和真实性,得到最优的目标跟踪结果。在OTB-2015和VOT-2017两个数据库中进行大量测试,试验结果表明,本研究所提方法与LSART算法相比鲁棒性提高了10%,准确性提高了3.9%,并且对目标遮挡和尺度变化具有出色的性能表现。

Due to the interference of various factors in the complex situation of reality,the trackers had some problems such as model drift and tracking failure.An adaptive fusion target tracking based on joint detection algorithm was proposed to improve the robustness and accuracy of the tracker.The deep and shallow convolutional features acted on the correlation filters separately to obtain response scores according to their respective advantages,and adaptively fused the response scores of different convolutional features by minimizing the loss.Then it combined with the location detection method to judge the validity and authenticity of the predicted location,so as to get the optimal target tracking results.A large number of tests were done in two open databases:OTB-2015 and VOT-2017.The experimental results showed that the proposed method was 10% more robust and 3.9% more accurate than the LSART algorithm.It also had excellent performance for occlusion and scale variation.