Objective Tracking learning Series 8: struck: Structured output tracking with Kernels (2011 iccv)

It seems that human-computer interaction is not as simple as I think, it still takes a lot of effort to lay the foundation. Then we will learn some tracking-related article algorithms.

After carefully studying the compression tracking (CT), I feel that I have a good understanding of tracking. However, after reading the results on the test set, the original CT effect is very poor on the real camera. The only advantage is fast.

Therefore, you have to come back to learn other methods with high precision! Struck is selected here. Recommended by Daniel senior. At the same time, this article uses the non-linear SVM algorithm and also wants to learn it by the way.

Position of the article code: http://www.samhare.net/research indicates I haven't entered yet, But I found somewhere else.

Let's take a general look at the innovation points of this paper:

1. SVM classification with accurate classification should be better than Bayesian and Adaboost, but the speed should not be good (we can improve it in this aspect );

2. Instead of directly outputting prediction labels, It outputs various transformations of the object. Then select the best transform from it;

3. For the training samples, either the CT or the object is very close or positive, or the long distance is negative. Instead, overlap is calculated. If it is greater than a certain threshold, it is regarded as a positive sample. If it is smaller than a certain threshold, it is considered as a negative sample. The tag in the middle is regarded as 0;

4. Use Gaussian Kernel functions;

5. The number of support vectors is limited. The support vector with the least impact on the weight will be discarded.

I want to check the details of this paper against the code of this paper. I hope this paper will help me know the basic processing methods of tracking with high precision. It is best for me to modify it in the framework that I am familiar with now. On the other hand, I can know how SVM is used in CV.

Furthermore, my ultimate goal remains the same. I Want To Do human-computer interaction! Welcome to join us!

Iker is constantly updated.

Objective Tracking learning Series 8: struck: Structured output tracking with Kernels (2011 iccv)