Is it possible to combine HoG and AdaBoost algorithms for tracking?

Question

Is it possible to combine two tracking algorithms namely HoG and AdaBoost?

Or are there any video tracking algorithms which can be combined? I'm trying to develop an algorithm by combining these two. Is it possible? Also I'm trying to implement it using OpenCV.

Answer 1

HoG is a feature(descriptor) extraction method, and Adaboost is a machine learning algorithm to detect (not track) object.

So, you can train the machine to detect the object by it's HoG descriptor using adaboost algorithm, then detect.

In a word, HoG + Adaboost is for detection, not tracking.

You can use particle filter, LK tracker or etc, for tracking.

Answer 2

AdaBoost is not exactly a face tracking algorithm - it's a method of increasing performance of any learning algorithms. HoG is a specific method of finding shapes and patterns on image (more exactly - describing those). You should read more on that topic:

https://www.cs.princeton.edu/~schapire/papers/explaining-adaboost.pdf - nice paper on AdaBoost (if you need something more quick and dirty - just look on wikipedia)

http://www.cvc.uab.es/~davidm/pdfs/IBPRIA2011.pdf - again wiki could be of much help for you.

There is a well-known technique of face detection which combines haar-like features recognition and Adaboost - see OpenCV library and CascadeClassifier class there (http://docs.opencv.org/modules/objdetect/doc/cascade_classification.html)