没有合适的资源?快使用搜索试试~ 我知道了~
Multi-view Multi-object Tracking Based on Global Graph Matching ...
0 下载量 70 浏览量
2021-02-22
09:42:22
上传
评论
收藏 1.42MB PDF 举报
温馨提示
试读
10页
Multi-view Multi-object Tracking Based on Global Graph Matching Structure
资源推荐
资源详情
资源评论
Multi-view Multi-object Tracking
Based on Global Graph Matching Structure
Chao Li, Shantao Ping
(&)
, Hao Sheng, Jiahui Chen, and Zhang Xiong
State Key Laboratory of Software Development Environment,
School of Computer Science and Engineering, Beihang University,
Beijing 100191, People’s Republic of China
{licc,pingst,shenghao,chenjh}@buaa.edu.cn
Abstract. We present a novel global graph matching framework based on
virtual nodes for multi-object tracking in multiple views. Contrary to recent
approaches, we incorporate a global graph matching structure (GGMS),
allowing the tracker to better cope with long-term occlusions and tracking
failure caused by interaction of targets. In our approach, the matching problem is
solved as follows: Virtual detections are introduced by mapping the nodes
among views, to ensure that the amount of detections in each view is the same,
and then realize the whole graph matching. In addition, appropriate optimization
is performed to convert this mapping problem to the Assignment Problem,
which could be efficiently addressed by the Hungarian Algorithm. Finally, we
demonstrate the validity of our approach on the publicly available datasets, and
achieve very competitive results by quantitative evaluation.
Keywords: Multi-object tracking
Multi-view Graph matching
1 Introduction
With the fast development of smart devices, numerous cameras lead to ubiquitous
video sources. Crowd-sourced video retrieval systems [1] based on video content
comparison has emer ged. Multi-object tracking is a key problem for many computer
vision tasks, such as surveillance [2], animation or activity recognition. The tracking in
video consists of detecting all subjects in every frame, and following their complete
trajectory over time. Successful research on a new generation of reliable pedestrian
detectors [3, 4] has prompted the use of the tracking-by-detection paradigm [5], even
for crowded or semi-crowded scenarios. Under this paradigm, the problem is often
divided in two steps: detection and data association [ 6 , 7]. The tracker first acquires a
set of detections using a pedestrian detector. The individual detections are then
assigned to tracks, where each track is composed of all the detections from a single
individual. If all persons were to be correctly observed at every timestamp this task
would be trivial, however, due to false positive detections, occlusions and missed
detections, this association problem becomes very challenging.
Usually, the problem of tracking is divided into two directions: monoc ular tracking
and multi-view tracking. In recent research, the minimum-cost network flow tracking
approach [8, 9] is more popular in monocular tracking. This method can effectively
© Springer International Publishing AG 2016
E. Chen et al. (Eds.): PCM 2016, Part II, LNCS 9917, pp. 660–669, 2016.
DOI: 10.1007/978-3-319-48896-7_65
资源评论
weixin_38516863
- 粉丝: 3
- 资源: 970
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功