基于孪生网络在视频目标跟踪中的应用

On Application of Twin Networks to Video Target Tracking

针对目标跟踪出现的遮挡、相似干扰、快速运动等难题,提出了在孪生网络跟踪框架基础上进行遮挡检测的RGB-D视频目标跟踪算法。将聚合多层次特征的深度可分离卷积孪生网络作为目标跟踪基准框架,实验结果表明,提出的视频目标跟踪算法能够稳定跟踪目标,具有较高的成功率。

Aiming at the problems of occlusion,similar interference and fast motion in target tracking,a new occlusion detection algorithm for RGB-D video target tracking based on Twin Networks Tracking Framework is proposed.The depth-separable convolution Twin Networks with multi-level features are used as the frame of reference for target tracking.The experimental results show that the proposed algorithm can track the target stably at a high success rate.