摘要
大多数现代的跟踪器的核心元素是一个判别式的分类器,这个分类器的任务是去区分目标区域和周围的环境。本文基于基础的相关滤波的方法,并对其进行改进的目的,通过在相关滤波的计算中引入不同的尺度信息,较好的手动设计的特征和更具表示性的深度特征,然后在OTB2013数据库上对这些不同的方法进行实验,得到基于深度特征和多尺度信息的方法比基础的相关滤波的方法的重叠精度(Overlap Precision,OP)提高了16.68%(从62.77%到79.45%),从而验证了这些额外信息的引入可以很大地提升跟踪性能。
The core component of most modern trackers is a discriminative classifier, tasked with distinguishing between the target and the surrounding environment. In this paper, we aim to improve the object tracking performance based on the basic correlation filter method. By introducing multi-scale information, better hand-crafted feature, representative deep feature and experimenting on OTB2013 benchmark, the method using multi-scale and deep feature gets 16.68% absolute performance improvement from 62.77% to 79.45% compared to the basic method. The experiment result verifies that object tracking performance can improve greatly by introducing extra information.
作者
董艳兵
DONG Yan-bing(Shanghai Institute of Microsystem and Information Technology,Chinese Academy of Sciences,Shanghai 200050,China;School of Information Science＆ Technoloy ShanghaiTech University,Shanghai 201210,China;University of Chinese Academy of Sciences,Beijing 101407,China)
出处
《电子设计工程》
2018年第20期6-9,13,共5页
Electronic Design Engineering
关键词
目标跟踪
相关滤波
判别式分类器
多尺度
深度特征
object tracking
correlation filter
discriminative classifier
multi-scale
deep feature
