CODEWITHSUNDEEP
pythonpython-2.7opencvtracking
Iwn
Iwn

Reputation: 65

Object tracking in video

I'm trying to tracking people in the video. But I can not find a suitable algorithm that would behave similarly to https://www.youtube.com/watch?v=Qjr3RYecv3U.

I tried template matching in combination with optical flow, but always lose the tracked object if it overlaps another object. Could someone recommend a suitable method for tracking?

I am using Python and OpenCV.

Upvotes: 0

Views: 2901

Answers (2)

Kousik Krishnan
Kousik Krishnan

Reputation: 98

If you are looking for some quick code which runs in CPU, take a look at Drew-NF. This is a python implementation of the neural network discussed in the paper Tubelets with Convolutional Neural Networks for Object Detection from Videos. To Run the script you need:

  1. Tensorflow

  2. OpenCV

DrewNF Github Repo

Upvotes: 1

Brian O'Donnell
Brian O'Donnell

Reputation: 1876

The results of the ILSVRC 2017 competition were released yesterday (July 17, 2017). The winner in the two tracking categories, Task 3c (Object detection/tracking from video with provided training data) and Task 3d (Object detection/tracking from video with additional training data), was this team:

Jiankang Deng(1), Yuxiang Zhou(1), Baosheng Yu(2), Zhe Chen(2), Stefanos Zafeiriou(1), Dacheng Tao(2), (1)Imperial College London, (2)University of Sydney

Here are their publications, source code, and a presentation: [1] Deep Feature Flow for Video Recognition Xizhou Zhu, Yuwen Xiong, Jifeng Dai, Lu Yuan, and Yichen Wei, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017.

[2] Flow-Guided Feature Aggregation for Video Object Detection, Xizhou Zhu, Yujie Wang, Jifeng Dai, Lu Yuan, and Yichen Wei. Arxiv tech report, 2017.

Presentation https://www.youtube.com/watch?v=J0rMHE6ehGw

Source Code https://github.com/msracver/Deep-Feature-Flow

The code has the following prerequisites:

  • Python 3.2.0+
  • Microsoft's MXNet
  • Cython
  • OpenCV (Python bindings)

Their code requires a GPU with at least 6GB of memory.

Another option is ROLO. The author is Guanghan Ning and he uses You Only Look Once (YOLO) for detection and uses TensorFlow to implement LSTMs for tracking.

His published a paper: Spatially Supervised Recurrent Convolutional Neural Networks for Visual Object Tracking, IEEE International Symposium on Circuits and Systems, 2017

His code is here: https://github.com/Guanghan/ROLO

Project page: http://guanghan.info/projects/ROLO/

Prerequisites:

  • Python 2.7 or 3.3+
  • TensorFlow
  • Scipy
  • OpenCV (Python bindings)

Some videos of his work:

Upvotes: 2

Related Questions