RGB-D Object Tracking and Occlusion Deformation Processing Based on Depth Model
首发时间:2019-04-18
Abstract:To achieve more accurate RGB-D tracking, robust occlusion and deformation processing, an object tracking method based on depth model is proposed. The algorithm is based on the kernelized correlation filter to satisfy the real-time requirement. The tracking situation is determined by the tracking results of kernelized correlation filter on the color image and depth image. In the unambiguous situation, the linear regression model is adapted to fuse the tracking results. In the ambiguous situation, the relative weight method is used to fuse the tracking results. The Gaussian mixture model is adapted to judge the target state. Partial occlusions and deformations are processed by the region growing method according to the target depth range determined by the model. The algorithm is evaluated using the Princeton data set. The experimental results demonstrate that the tracker achieves more accurate and robust tracking results when the target is partially occluded or deformed.
keywords: RGB-D tracking deformation processing occlusion processing kernelized correlation filter Gaussian mixture model region growth method
点击查看论文中文信息
基于深度模型的RGB-D目标跟踪及遮挡与变形处理
摘要:为了实现更精确的RGB-D目标跟踪及稳健的遮挡与形变处理,提出一种基于深度模型的目标跟踪方法。该算法为保证实时性,采用核化相关滤波器进行构建。根据彩色与深度图像并行核化相关滤波器跟踪结果划分跟踪状态,在确定状态下,采用线性回归模型融合跟踪结果;在复杂状态下,采用相对权值法融合跟踪结果,利用高斯混合模型判断目标状态,根据模型确定的目标深度范围采用区域生长法处理部分遮挡或形变。采用普林斯顿数据集对该算法进行评估,结果表明,当目标发生部分遮挡或形变时,跟踪器能够取得更精确、稳健的跟踪结果。
引用
No.****
同行评议
勘误表
基于深度模型的RGB-D目标跟踪及遮挡与变形处理
评论
全部评论0/1000