多时空感知相关滤波融合的目标跟踪算法

Target Tracking Algorithm Based on Multi-Time-Space Perception Correlation Filters Fusion

为提高不同复杂环境下视频目标跟踪的可靠性,结合目标和背景的时空特性,提出一种基于多时空感知相关滤波融合的目标跟踪算法.该算法以相关滤波为基础,通过计算目标和滤波器在帧间变化的一致性,从而确立时间特性;以跟踪目标的邻域信息为基础,引入掩模矩阵,提取目标的空间信息.最后在目标函数中,构建时空感知约束项,强化相关滤波器对时空二元信息的学习能力,进而增强了滤波器对干扰信息的鲁棒性.为增强对不同复杂环境的适应能力,在视频颜色和方向梯度特征空间中搭建独立的时空感知相关滤波器,并建立二者跟踪结果的自适应融合机制,准确计算目标的位置和尺度,从而有效地提升算法对不同复杂环境的泛化能力.为验证算法的有效性,在OTB标准数据集上采用精度图和成功率图作为评价与11种算法开展了对比实验.实验结果表明,所提算法在多种复杂环境下,对遮挡、形变、光照变化和快速移动等干扰均具有良好的鲁棒性,并对目标能够进行有效的跟踪.

In order to improve the reliability of video target tracking under different complex environments,this paper proposes a target tracking fusion algorithm based on multi-time-space perception correlation filters,which combines the time and space characteristics of target and background.The time characteristic is established by calculating the consistency of changes between target frames and filter frames based on the correlation filter.The target spatial information is extracted from the mask matrix and the neighborhood of the target.Finally,in the objective function,the constraints term of time-space perception is introduced to enhance the learning ability of correlation filter to improve the robustness of interfering information.In order to improve the ability of adapting to a complex and changeable environment,time-space perception correlation filters are established respectively in the color space and orientation gradient space.Then an adaptive fusion mechanism of the two tracking results is established to accurately calculate the target position and scale,which effectively improves the generalization ability of the algorithm for different complex environments.To validate the effectiveness of the proposed algorithm,comparison experiments with other 11 algorithms were performed on the OTB standard data set.The experimental results show that the algorithm in this paper has good robustness for occlusion,deformation,illumination variation,fast motion and other disturbances in a variety of complex environments,and can effectively track the target.