TLD long-term tracking algorithm
Published:
The main approach taken in the TLD algorithm is the concept of online training. Since the object is unknown until runtime, training can not be decoupled from detection. The algorithm utilizes a short-term tracker working in tandem with a sliding-window detector to train a random forest classifier on-the-fly, and grow and improve the decision forest with time.
The project was an implementation of the original work by Zdenek Kalal.