针对繁杂环境下目标跟踪稳定性差且易受到遮挡发生漂移的问题, 提出一种结合时空上下文信息的相关滤波目标跟踪方法。该算法首先从目标和背景区域提取方向梯度直方图特征和颜色直方图特征, 加权融合两种特征的相关滤波响应, 建立相关滤波跟踪模型; 然后利用目标的背景梯度直方图特征, 基于贝叶斯框架通过最大化似然函数得到时空上下文辅助模型; 最后自适应融合两种模型响应, 得到目标估计位置并采用尺度估计方法解决目标尺度变化问题。在 OTB-2013公开标准测试集上与基于相关滤波的运动目标跟踪方法进行了实验对比。结果表明, 该算法的平均距离精度值和平均重叠精度值都优于其他算法, 能够有效缓解跟踪目标由于遮挡、尺度变化、光照等因素造成的跟踪漂移状况的发生。

Target tracking is poor in complex environments and is vulnerable to occlusion drift; therefore, a filtering target tracking method using spatiotemporal context information is proposed here. First, the algorithm extracts the direction gradient histogram feature and the color histogram feature from the target and background regions. Then, the correlation filter response of the two features are weighed to establish the correlation filter tracking model. The background gradient histogram feature of the target is used, based on Bayesian. The framework obtains the spatio-temporal context-assisted model by maximizing the likelihood function. Finally, the two models are adaptively combined to obtain the target estimation position, and the scale estimation method is used to solve the target scale change problem. The correlation-based moving target tracking method was experimentally compared on the OTB-2013 open standard test set. The results show that the average distance precision value and average overlap precision value of the algorithm were better than other algorithms, which can effectively address the tracking drift caused by occlusion, scale change, and illumination.