基于时空上下文模型的RGB-D序列目标跟踪方法

Object Tracking in RGB-D Sequences Using a Spatio-Temporal Context Model

为了实现更为精确的视频目标跟踪,提出一种以时空上下文模型为基础的RGB-D序列目标跟踪算法.通过引入更新模板的深度信息,该模型精准地区分了输入序列的目标区域与背景区域,实现了深度权值和颜色权值的融合;基于目标序列的深度及目标动量计算,该模型有效地实现了尺度更新与遮挡处理.通过在RGB-D图像序列数据集上的详细实验评估,该时空上下文模型相对于其他先进的同类方法表现出更好的性能.因此,该方法实现了更为精确可靠的视频目标跟踪.

To improve the precision of object tracking in videos,this work presented an RGB-D tracking method using a spatial-temporal context(STC)model.By introducing depth data,STC can clearly distinguish target from background in the context,and perform effective fusion of the depth weights and color weights.At the same time,based on the depth information and the target momentum,the proposed method is capable of adjusting scale and handling occlusions.As a result,the proposed tracker is able to produce precise prediction of target locations even when the target object is under severe occlusion.Comprehensive evaluations on challenging datasets demonstrate that the proposed tracker gives favorable performance over several state-of-the-art counterparts.Consequently,the proposed method in this work is capable of achieving more precise and reliable object tracking in videos.