通道加权和自适应模型更新的实时目标跟踪

针对单一特征不能有效建模目标及特征通道直接叠加影响跟踪质量的问题，基于判别式尺度空间跟踪（DSST）提出了自适应模型更新与响应加权的实时相关滤波跟踪算法。首先，融合多维特征，并对每一响应通道自适应加权；其次，在尺度估计中融入多项式拟合策略；最后，根据响应图峰值波动情况进行样本可靠性判定，并提出模型更新策略。在OTB50和OTB100上进行实验，所提算法相较基准跟踪器，成功率分别提升6.4%和6.7%。与最近的跟踪器相比，其展现了优秀的性能，每一模块都对结果产生了有效的提升，且速度高达85 fps，是基准跟踪器DSST的三倍，超过大部分实时性跟踪算法。

Real-time object tracking algorithm with adaptive model updates and channel weighting

Aiming at the problem that single feature cannot effectively model targets and the direct superposition of response channels affect the tracking quality. Based on discriminative scale space tracking(DSST), this paper developed an adaptive model update and response weighted real-time correlation filter tracking algorithm. Firstly, the algorithm integrated multi-dimensional feature, and applied adaptive weighting to every response channel. Secondly, the proposed tracker incorporated the quadratic polynomial fitting strategy to improve the accuracy of scale estimation. Finally, it introduced the model update strategy to improve the reliability of the sample and accelerate the tracker. It performed experiments on OTB50 and OTB100. Compared with the baseline tracker, the proposed algorithm improves 6.4% and 6.7% in success. Compared with the recent tracking algorithm, it shows excellent performance, and has a speed of 85 fps, which is three times than that of DSST, and exceeds most real-time tracking algorithms.