基于置信度的加权特征融合相关滤波跟踪

Weighted Feature Fusion Correlation Filter Tracking Based on Confidence Level

目前机器视觉应用广泛,视频目标跟踪的过程中会遇到各种挑战。为解决单一特征鲁棒性差,模型和尺度更新机制不健全的问题,提出了一种将自适应加权特征融合方法与置信度模型及尺度更新机制相结合的相关滤波目标跟踪算法。算法将互补的梯度和颜色特征进行融合,通过计算各特征滤波响应来决定下一帧在融合特征中各自所占的权重,凸显优势特征,使目标与背景更具区分度。同时引入置信度更新机制,防止模型更新引入遮挡物、相似干扰,提高正确率。最后提出一种新的尺度更新策略,简化冗余代码,使跟踪更精确的同时降低时间代价。实验结果证明,该算法在精度和正确率上都比几种现有相关滤波算法更优,应对相似目标干扰和遮挡情况具有更高鲁棒性。对相关滤波算法进行了改进,加入了特征融合和更新机制,使算法提高了跟踪效果,具有一定的应用价值。

Nowadays, the machine vision is widely used, and there will be various challenges in the process of video target tracking. In order to solve the problem of weak robustness, model and scale update mechanism, a correlation filter target tracking algorithm that combines the adaptive weighted feature fusion method and the confidence degree model updating mechanism is proposed. The algorithm uses complementary gradient and color features for feature fusion. By calculating the filtering response of each feature, it determines their weight in the fusion feature in next frame, so as to highlight the dominant feature and make the target more distinguishable from the background. At the same time, the confidence level is introduced. The peak side lobe ratio is used as the evaluation criterion to prevent model updating from causing similar interference and occlusion, which can improve accuracy. Finally, a new scale update method is proposed, which simplifies the redundant code to reduce the time cost as well as make the tracking more precise. Experimental results show that the proposed algorithm is better in accuracy and success rate than several existing correlation filtering algorithms, and it is more robust to deal with similar target interference and occlusion. This paper improves the correlation filtering algorithm, the feature fusion and updating mechanism is added, so that the algorithm improves the tracking effect and has certain application value, it has certain application value.