Main idearTreat the tracking problem as a classification task and use online learning techniques to update the Object Model
Main innovative points1 Based on structured output prediction (structured SVM), in which the task is directly predict the change in object location between frames, instead of relying on a heuristic intermediate step for producting labeled binary samples with which to update the classifier, which is often a source of error during tracking
2 The online structured output SVM learning framework is also easy to induplicate ate image features and kernels, and SVM also has good generalization ability, robustness to label noise and flexibility in object representation through the use of kernels
3. The positive sample is not near the target, but overlap is used for the positive side of the negative sample.
4-core functions
5. Restrict the support vector and remove the smaller impact of the weight.
The issues raised by other tracking-by-detection approach1 The classification confidence function provides an accurate estimate of object position is
Not explicitly inreceivated into the Learning Algorithm, Since the classifier is trained only with binary labels and has no information about Transformations
2 examples used for training the classifier are all
Equally weighted, Meaning that a negative example which overlaps significantly with the tracker bounding box is treated the same as one which overlaps very little. one implication of this is that slight inaccuracy during tracking can lead to poorly labeled examples, which are likely to reduce the accuracy of the classifier, in turn leading to further tracking Inaccuracy
3 The labeller is
Usually chosen based on intuitions and Heuristics, Rather than having a tight coupling with the classifier. mistakes made by the labeller manifest themselves as label noise, and keep current state-of-the-art approaches try to overcome this problem by using robust loss functions [13, 14], semi-supervised learning [11, 17], or multiple-instance learning [3, 23]. we argue that all of these techniques, though justified in increasing the robustness of the classifier to label noise, are not addressing the real problem which stems from separating the labeller from the learner
How to do1. The overview algorithm consists of two steps: 1. Estimate the displacement of an object and 2. Update the discriminant function.
Structure Learning is a learning method that allows the output to be structured. Theoretically, any output can be used as a structure to solve any problem. Structure SVM is a type of structure learning, it is a mature Algorithm Implementation Framework. For more information, see Appendix 1, 2, and 3.
2. The model in the structure SVM model is: where the constraints are 1> = 0 evolved from> = 0, the solution of W is not unique, therefore, we select the W with the maximum interval and limit the W length. --> = 1 2 relaxation> = 1-3: replace 1 with the loss function. The more different, the stricter the interval requirement (larger)
3. Solve this SVM model
The a core step is based on the SMO (minimum Sequence Optimization) style. The basic idea of SMO is to select two variables (at least one violation of kkt) and fix other variables, two variable quadratic programming problems are solved. This way, the problem is constantly decomposed into subproblems for solving, and thus the goal of solving the original problem is achieved. SMO reference 4, 5
In section B, budget is introduced to constrain the number of support vectors, so as to achieve the real-time method: similar to [21]. we choose to remove the support vector which results in the smallest change to the weight vector
W, As measured
|?
W|2
C search over y on a polar grid rather than considering every pixel offset.
Lab1 uses 2 scales of 6 different Haar-like (192 features) 4x4 2 combine some different features by averaging multiple kernels
Reference attachment1 Large Margin Method for structured learning2 Support Vector Machine Learning for interdependent and structured output spaces 1 Short Version 3 structured learning and prediction in computer vision4 statistical learning method-Li Hang 7.45 sequential minimal optimization: A Fast Algorithm for Training Support Vector machines6 thesis homepage (including code)
Struck: structrued output tracking with kernels paper note
