基于特征融合和尺度自适应的核相关滤波算法研究

Research on Kernel Correlation Filtering Algorithm Based on Feature Fusion and Scale Adaptation

针对传统核相关滤波算法中存在单一特征鲁棒性差、尺度更新不健全等问题,提出了一种特征融合和尺度自适应相结合的目标跟踪算法。首先,通过研究各特征的侧重,将方向梯度特征(HOG)和颜色特征(CN)相融合,可以得到一种辨别能力更强的特征。然后,加入BRISK特征目标尺度的预测方式,从而实现目标的尺度更新,进而提高目标的跟踪精度。最后,在OTB2013数据集上进行算法对比实验。结果表明,与传统算法相比,改进后的算法不仅对于目标遮挡、光照变化和运动模糊等问题鲁棒性较高,而且满足实时跟踪需求。

Aiming at the problems of poor robustness of single feature and imperfect scale updating in traditional kernel correlation filtering algorithm,a target tracking algorithm based on feature fusion and scale adaptive is proposed.Firstly,by studying the emphasis of each feature,the Histogram of Oriented(HOG)feature and Color Names(CN)feature are combined to get a feature with stronger discrimination ability.Then,the prediction method of BRISK feature target scale is added to update the target scale and improve the target tracking accuracy.Finally,the algorithm comparison experiment is carried out on OTB2013 data set.The results show that,compared with the traditional algorithm,the improved algorithm not only has higher robustness to the problems of object occlusion,illumination change and motion blur,but also meets the requirements of real-time tracking.