Reproduced TLD (tracking-learning-detection) target tracking

Objective:

Original address: http://www.cnblogs.com/lxy2017/p/3927456.html

Brief introduction:

Original: http://blog.csdn.net/mysniper11/article/details/8726649

Video Introduction Website: http://www.cvchina.info/2011/04/05/tracking-learning-detection/

TLD (tracking-learning-detection) is a new single-target long term Tracking tracking algorithm presented by Zdenek Kalal, a Czech PhD student at the University of Surrey, UK, during his doctoral degree. The significant difference between the algorithm and the traditional tracking algorithm is that the traditional tracking algorithm and the traditional detection algorithm can be used to solve the deformation and partial occlusion of the tracked target during the tracking process. At the same time, through an improved on-line learning mechanism, the "salient feature point" of tracking module and the target model of detecting module are updated continuously, which makes the tracking effect more stable, robust and reliable.

A key issue for long-term tracking is that the system should be able to re-detect it and start re-tracking when the target is re-appearing in the camera's field of view. However, in the course of long-term tracking, the target will inevitably change shape, illumination condition, scale change, occlusion and so on. The traditional tracking algorithm, the front-end needs to match with the detection module, when the tracked target is detected, then began to enter the tracking module, and thereafter, the detection module will not be involved in the tracking process. However, this method has a fatal flaw: that is, when the tracked target has a shape change or occlusion, the tracking is easy to fail, so for long-time tracking, or the tracking target has the shape of the changes in the case of tracking, many people use detection method to replace the tracking. Although this method can improve the tracking effect in some cases, it requires an offline learning process. That is, before testing, a large number of samples of the tracked targets need to be selected for learning and training. This means that training samples will cover the various deformations and variations in scale, posture and illumination that may occur in the tracked target. In other words, the use of detection method to achieve long-time tracking purposes, training samples for the choice of the most important, otherwise, the robustness of the tracking is difficult to guarantee.

We recommend that you visit the foreign author's homepage: http://info.ee.surrey.ac.uk/Personal/Z.Kalal/

On the author's website there are downloads of relevant source code and some useful papers for him:

Original author of the source code for MATLAB and C mixed programming, has already someone to share the C + + version of the source: http://gnebehay.github.com/OpenTLD/

The source code on GitHub can also be used: Https://github.com/arthurv/OpenTLD

The more useful blogs on csdn are:

(1) "Discovering TLD" series:

http://blog.csdn.net/yang_xian521/article/details/7091587

(2) "Re-talk about PN learning":

http://blog.csdn.net/carson2005/article/details/7647519

(3)--TLD tracking algorithm for video tracking algorithms that are stronger than Microsoft Kinect

http://blog.csdn.net/carson2005/article/details/7647500

(4) "TLD Visual tracking Technology Analysis"

Http://www.asmag.com.cn/number/n-50168.shtml

(5) "TLD (tracking-learning-detection) Learning and source Understanding (i)"

http://blog.csdn.net/zouxy09/article/details/7893011

It is hoped that this information will be helpful to those who wish to learn the TLD algorithm.

From: http://blog.csdn.net/windtalkersm/article/details/8018980

TLD is the abbreviation of an algorithm, which is called tracking-learning-detection by the original author. Visual people see this name will be startled, very ambitious plan. It was a 09 job, not too long, but not too new. There are a lot of resources on the internet about this, much to do with the author's Open source code.

The first problem encountered in the learning process is that there are too many resources---of course it is relative to this field, generally can find a faithful reproduction algorithm of the source code is already very good. So to find the list, although a bit of a waste of time, I hope that can be helpful to others. Specific details are not much to say, there are a lot of great analysis is listed below, such as zouxy09 write the source code comments, can not be more detailed. If hard to pick a fault, it is a large paragraph of the text makes people dizzy, also did not how typesetting. I would like to draw a few simple pictures to add, I do not know what good point of the drawing program recommended (latex, or GNUPlot?)

Source Code Resources:

1. Original author Zdenek Kalal

Author's homepage: http://info.ee.surrey.ac.uk/Personal/Z.Kalal/

Source code page: Https://github.com/zk00006/OpenTLD

Programming language: Matlab + C

2. Alan Torres Edition

Source code page: Https://github.com/alantrrs/OpenTLD

Implementation language: C + +

3. Version Arthurv

Source code page: Https://github.com/arthurv/OpenTLD

Implementation language: C + +

Note: And none of the above found any difference

4. Version Jmfs

Source code page: Https://github.com/jmfs/OpenTLD

Implementation language: C + +

Note: There is no difference from the above two, only joined the VS2010 project file, in theory can be directly under the Windows compiler. But OpenCV can't detect the author's webcam (!!!) ), so he used another Videoinput class to handle the camera input.

This is a adaptation of arthurv ' s fork of Opentld (Https://github.com/arthurv/OpenTLD) to being immeadiately runnable in Vis UAL Studio 2010.

5. Georg Nebehay Version (finally a different one ....) ）

Source code page: http://gnebehay.github.com/OpenTLD/

Note 1: The benefit of this is to provide executable file downloads (Ubuntu 10.04 and Windows). But, as you would expect, basically on your machine can not run. Just build it yourself.

Note 2: This version requires QT installation. But it seems that the author turned off the QT option (the associated code is still there), so it compiles, but the results are not displayed.

Note 3:CSDN download has a "opentld Qt version", this is the version. But adding a project file for vs---still not pnp on my machine, don ' t bother

http://download.csdn.net/download/muzi198783/4111915

6. Paul Nader Edition (another QT version!) ）

qopentld:http://qopentld.sourceforge.net/

Source code page: http://sourceforge.net/projects/qopentld/

Note 1: OpenCV and QT are required. The original system requires QT 4.3.7OpenCV 2.2.

Note 2:windows and Linux are provided with either a compile project or a makefile. It is also the only tld! to be ported to the Android platform.

7. Ben Pryke Edition (another student project! ）

Source code page: Https://github.com/Ninjakannon/BPTLD

Note: The matlab+c/c++ is still a hybrid implementation. The highlight is a very detailed documentation (8 pages), which introduces the understanding and implementation details of the algorithm. can help understand the original algorithm

What you want to say:

1. Share: The previous period of time to Tld::init (...) After reading, Ben wanted to bang, other things too much had to put down. However, I am more familiar with detection and tracking, Init has already put the learning done again, understand the rest is easy. Now re-pick up, inadvertently found the Zouxy09 notes, save too much strength, half a day to see the finished. A lot of details do not have to take their own---we often complain that the resources of the document is too few, envy foreigners ability strong, quick action, and they are willing to share too much. Often see some good collection of articles, a few days to see incredibly deleted!

It may be understood that the company should be kept secret, but if the fear that others know their own ideas can not be done, it is still not in this area to do. The algorithm is only thought, and no one can monopolize it. The algorithm must be constantly updated, holding an algorithm will not survive a few years. The original author also started the company based on this technology, and did not see them based on this restriction others use. Sift,surf are patent and haven't heard of making a lot of money, Kinect tells you the algorithm can not be realized. The secret is to implement the details

2. Comparison: Finally read the realization, the overall feeling of the algorithm is more like engineering applications (engineering) rather than theoretical breakthroughs (also can not ask too much is not). It is not necessarily better than a single tracking (tracking) or detection (detection-by-classification) module, after all, that the two fundamental challenges of appearance (appearence) and scale change are not resolved. But this kind of framework should actually be very practical, because----------adjustable parameters too much!

TLD believes that many people have tried, real-time many people are complaining, and get their own video on the total need to adjust the parameters of the effect is good.

More like this year ECCV Kaihua Zhang's compressive Tracking: Theoretical advanced Scary (joking), the source of simple scary. And so far I have tried the off-the-shelf tracker in the best tracking effect, do not adjust any parameters, absolute real-time----code so little, want to not real-time is difficult (by the way, the author's blog mentioned above). This is the way to do research, there is a strong theory to support the implementation can be very simple without affecting the effect. So if the math people are willing to do the application, a lot of people will be laid off

Http://www4.comp.polyu.edu.hk/~cslzhang/CT/CT.htm

Another PWP (Pixel-wise posteriors), publish time and TLD, looks pretty good, but the author says it's open source and hasn't been cashed. is a pity. Individual feel that level set on partial occlusion effect should be very good, real-time is not difficult

Http://www.robots.ox.ac.uk/~cbibby/research_pwp.shtml

3. Summary: TLD is actually a very suitable entry and advanced algorithm:

A. Theory, high-quality paper (BMVC, CVPR, ICPR, last Pami)

B. Active Code! Matlab, C + +, Windows, Linux, ..... What else do you want?

C. A detailed introduction and detailed code comments that are shared by different bull calves (almost every line is explained)!

4. Involved in a wide range, involving detection, tracking, classifcation, traditional visual technology is so stiff division of the three major categories. After the study of each part of how much can be a bit of experience.

The End

TLD Source Understanding

TLD (tracking-learning-detection) Learning and source Understanding (i)

TLD (tracking-learning-detection) Learning and source Understanding (II.)

TLD (tracking-learning-detection) Learning and source Understanding (III.)

TLD (tracking-learning-detection) Learning and source Understanding (iv)

TLD (tracking-learning-detection) Learning and source Understanding (V)

TLD (tracking-learning-detection) Learning and source Understanding (vi)

TLD (tracking-learning-detection) Learning and source Understanding (vii)

Reproduced TLD (tracking-learning-detection) target tracking