光学学报, 2020, 40 (17): 1715002, 网络出版: 2020-09-02
具有旋转特性的目标跟踪算法 下载: 818次
Target Tracking Algorithm with Rotation Characteristics
机器视觉 目标跟踪 极坐标系 目标旋转 傅里叶-梅林 machine vision object tracking polar coordinates target rotation Fourier-Merlin
摘要
目标发生旋转和尺度变化等时会导致跟踪算法出现目标丢失和精度大幅度下降等问题。因此解决目标在运动过程中出现的旋转以及尺度变化问题成为当前的研究热点。提出具有旋转特性的目标跟踪算法,该算法以Hamed等提出的 BACF(background-aware correlation filter)为基准,保留BACF算法中的定位,将笛卡儿坐标系下的目标特征转换到极坐标系下,并采用傅里叶-梅林公式来计算目标旋转角度和尺度的改变,在公开数据集POT上进行验证和比较,发现经过改进后的算法在目标旋转时,矩形框可以跟随目标发生旋转,并且本文算法在POT数据集上的准确率和成功率具有大幅度的提升,分别为0.6561和0.5930,旋转特性准确率和成功率分别为0.9619和0.8527。
Abstract
The target rotation and scale change lead to the target losing and precision decrease greatly. Therefore, it has become a hot topic to solve the rotation and scale change of the target in the processing of moving. This paper presents a target tracking algorithm with rotation characteristics. Based on the BACF (background-aware correlation filter) proposed by Hamed et al., the proposed algorithm retains the localization in BACF algorithm, and converts the target features under Cartesian coordinates into that under polar coordinates. The Fourier-merlin formula is used to calculate the changes in rotation angle and target scale, and the dataset POT is used to verification and comparison. It is found that, in the improved algorithm, the rectangle box can rotate with the target when the target is rotating. The accuracy and success rate of the algorithm in this paper are greatly improved on the POT dataset, which are 0.6561 and 0.5930, respectively, and the rotation characteristic accuracy and success rate are 0.9619 and 0.8527, respectively.
瑚琦, 李锐, 张薇. 具有旋转特性的目标跟踪算法[J]. 光学学报, 2020, 40(17): 1715002. Qi Hu, Rui Li, Wei Zhang. Target Tracking Algorithm with Rotation Characteristics[J]. Acta Optica Sinica, 2020, 40(17): 1715002.