摘要
当前的背景感知相关滤波跟踪方法通常采用固定的尺度步长与特征维度来估计目标的状态,导致跟踪精度和效率较低。针对该问题,本文提出基于随机尺度高效响应的背景感知目标跟踪方法。首先研究随机尺度步长估计,在视频每一帧随机生成多个不同尺度步长的目标搜索区域。然后探索多元特征高效响应,在单个目标搜索区域高效融合多个特征维度降低的响应图。最后实验结果表明,应用所提出的方法后,跟踪目标的精确度和成功率明显增加,运行速度提高约29.18%,具有良好的鲁棒性和实时性。
Current background-aware correlation filter-based tracking methods usually employ the fixed scale strides and feature dimensions to estimate the target state,resulting in low tracking accuracy and efficiency.Aiming at this problem,this paper proposes a background-aware target tracking method based on efficient responses with random scales.Firstly,we study the random scale stride estimation,which randomly generates multiple target search areas with different scale strides in each frame of the video.Then we explore the efficient responses with multiple features,which efficiently fuse multiple response maps produced by dimensionality reduction features in a single target search area.Finally experimental results show that after applying the proposed method,the precision and success of object tracking are significantly increased,and the running speed is increased by about 29.18%,which has good robustness and real-time performance.
作者
李胜杰
LI Shengjie(School of Computer Science(National Pilot Software Engineering School),Beijing University of Posts and Telecommunications,Beijing 100876)
出处
《软件》
2023年第2期1-6,共6页
Software
基金
中国博士后科学基金第70批面上资助项目“多重记忆协同感知的深度粒子滤波自适应跟踪方法研究”(2021M700516)。
关键词
目标跟踪
相关滤波
背景感知
随机尺度
高效响应
object tracking
correlation filter
background-aware
random scales
efficient responses
