多特征自适应融合的相关滤波目标跟踪算法

Multi-feature Adaptive Fusion Based Target Tracking Algorithm

本文针对单一特征目标相关滤波算法因光照变化、目标遮挡、低分辨率和运动模糊等导致目标跟踪的稳定性较差的问题,提出了一种将多种特征进行自适应融合的跟踪算法。本文算法在FDSST算法的基础上,自适应融合梯度直方图特征HOG(Histogram of Oriented Gradient)、颜色名特征CN(Color Name)和灰度特征来增强特征的表达能力;提出遮挡判断策略,能够有效的判断跟踪过程中的目标遮挡现象;引入目标重定位机制,在发生目标遮挡或干扰时,能够重新定位目标位置,有效的抑制跟踪漂移现象的产生。最后,本文选取OTB50和OTB100作为实验数据集,将本文算法和选取的六种主流算法进行性能比较。实验结果表明,本文算法在光照变化、运动模糊和目标遮挡等情况下的表现具有较高的稳定性和准确性;在成功率和跟踪精确度上都优于其他六种算法。

Aiming at the problem of poor target tracking stability caused by single-feature target-related filtering algorithms due to illumination changes, target occlusion, low resolution, and motion blur, this paper proposes a tracking algorithm that adaptively fuses multiple features. Based on the FDSST algorithm, the algorithm in this paper adaptively fuses gradient histogram features, color name features and grayscale features to enhance the expressive ability of features;proposes an occlusion judgment strategy, which can effectively determine the occlusion phenomenon of targets in the tracking process;introduce targets The relocation mechanism can relocate the target position when the target is blocked or interfered, effectively suppressing the occurrence of tracking drift. Finally, this paper selects OTB50 and OTB100 as the experimental data sets, and compares the performance of the algorithm proposed in this paper with the selected six mainstream algorithms. The experimental results show that the algorithm in this paper has high stability and accuracy under the conditions of illumination changes, motion blur and target occlusion;it is better than the other six algorithms in success rate and tracking accuracy.