[论文笔记]Real-Time* Multiple Object Tracking (MOT) for Autonomous Navigation
2018-03-04 17:00
1081 查看
论文地址: http://cs231n.stanford.edu/reports/2017/pdfs/630.pdf 1、摘要
本文提出了一种实时的多目标跟踪算法multiple object tracker (MOT)。先用Faster R-CNN进行检测,然后采用GOTURN (Generic Object Tracking Using Regression Networks)做跟踪。
![](https://oscdn.geek-share.com/Uploads/Images/Content/201803/f6893da00ea8c8532fa66a568cf1f341)
2. Related Work
3. Our Approach
![](https://oscdn.geek-share.com/Uploads/Images/Content/201803/c9a0476a2c85a814da03e75845d05ec4)
其中,GOTURN 实际上是单目标跟踪算法,主要思想是匹配相邻帧特征相似的位置,并回归坐标出来。
4. Dataset and Features
MOT challenge data set :每个视频相邻帧可以组 pair 训练
5. Method
训练时检测网络参数固定,只训tracking network。将检测结果 crop 出来( 宽w,高h),在下一帧 (2*w,2*h)的区域内回归坐标。
。
![](https://oscdn.geek-share.com/Uploads/Images/Content/201803/7423a91be510bf2c0a4229a64d3e545f)
6. Results
![](https://oscdn.geek-share.com/Uploads/Images/Content/201803/72904ad65b4990004a547f285cf8b123)
整体效果看起来一般
本文提出了一种实时的多目标跟踪算法multiple object tracker (MOT)。先用Faster R-CNN进行检测,然后采用GOTURN (Generic Object Tracking Using Regression Networks)做跟踪。
2. Related Work
3. Our Approach
其中,GOTURN 实际上是单目标跟踪算法,主要思想是匹配相邻帧特征相似的位置,并回归坐标出来。
4. Dataset and Features
MOT challenge data set :每个视频相邻帧可以组 pair 训练
5. Method
训练时检测网络参数固定,只训tracking network。将检测结果 crop 出来( 宽w,高h),在下一帧 (2*w,2*h)的区域内回归坐标。
。
6. Results
整体效果看起来一般
相关文章推荐
- 论文笔记 You Only Look Once: Unified, Real-Time Object Detection
- 论文阅读笔记:You Only Look Once: Unified, Real-Time Object Detection
- 【计算机视觉】【神经网络与深度学习】论文阅读笔记:You Only Look Once: Unified, Real-Time Object Detection
- A Twofold Siamese Network for Real-Time Object Tracking 阅读笔记
- 论文笔记| Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
- 论文阅读笔记:You Only Look Once: Unified, Real-Time Object Detection
- 论文笔记|You Only Look Once: Unified, Real-Time Object Detection
- 论文阅读笔记:You Only Look Once: Unified, Real-Time Object Detection
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks----论文笔记
- 多目标跟踪(MOT)论文随笔-POI: Multiple Object Tracking with High Performance Detection and Appearance Feature
- 论文阅读笔记:You Only Look Once: Unified, Real-Time Object Detection
- 论文笔记:目标追踪-CVPR2014-Adaptive Color Attributes for Real-time Visual Tracking
- [深度学习论文笔记][Object Detection] You Only Look Once: Unified, Real-Time Object Detection
- 论文笔记 You Only Look Once: Unified, Real-Time Object Detection
- 论文阅读笔记:You Only Look Once: Unified, Real-Time Object Detection
- 论文阅读-Re 3 : Real-Time Recurrent Regression Networks for Object Tracking
- Real-Time Compressive Tracking 论文笔记
- 论文笔记:目标追踪-CVPR2014-Adaptive Color Attributes for Real-time Visual Tracking
- 多目标跟踪(MOT)论文随笔-SIMPLE ONLINE AND REALTIME TRACKING WITH A DEEP ASSOCIATION METRIC (Deep SORT)
- 论文笔记:Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks