一种长宽比自适应变化的目标尺度估计算法

针对视觉跟踪中由于尺度变化、遮挡等复杂场景造成的跟踪失败问题,提出一种尺度长宽比自适应变化的目标尺度估计算法.该算法采用35×35个尺度因子来实现对目标的长宽比估计,为了降低运算量,通过分层尺度估计对二维尺度采样因子进行选择,既确定了目标的最佳尺度,又提高了算法的运行速度;为了进一步提高跟踪算法的鲁棒性,使用相邻两帧之间响应向量的欧式距离作为评判模板是否更新的标准.将尺度估计和模板更新模块引入到目前3种性能出色的相关滤波算法DSST、HCF和OSA中,进行仿真验证.实验结果表明,与原始算法相比,引入模块的新算法在跟踪成功率和精度上均有显著提高,在OTB100数据集上,成功率与3种原始算法相比,分别提高了1.3％、1.4％ 和1.4％,精度分别提高了1.2％、1.3％ 和1.0％,尤其在尺度变化、目标遮挡等复杂场景下具有明显的优势.

An Algorithm of Target Scale Estimation Based on Adaptive Change of Aspect Ratio

Aimed at the problems that tracking failure remains in visual tracking, being caused by complex scenes such as scale changes and occlusion, a target scale estimation algorithm with adaptive change of scale aspect ratio is proposed. An aspect ratio estimation of the target is realized by adopting 35×35 scale factors in the algorithm. In order to reduce the amount of calculation, the two dimensional scale sampling factor is selected by hierarchical scale estimation. By so doing, not only the optimal scale of the target is determined, but also the algorithm is improved. In order to further improve the robustness of the tracking algorithm, the Euclidean distance of the response vector between two adjacent frames is used as the criterion for judging whether the template is updated. The scale estimation and template update module proposed in this paper are introduced into the current three excellent correlation filtering algorithms DSST, HCF and OSA. The experiment results show that compared with the original algorithm, the tracking success rate and accuracy are improved significantly by using the new algorithm. On the OTB100 data set, the success rate increases by 1.3% ,1.4% and 1.4% respectively compared with the above three original algorithms. The accuracy increases by 1.2%, 1.3% and 1.0% respectively, especially in complex scenes such as scale changes and target occlusion. The new algorithm is advantageous to the visual tracking.