C. Huang, Y. Li and R. Nevatia, "Multiple Target Tracking by Learning-Based Hierarchical Association of Detection Responses," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 35, no. 4, pp. 898-910, April 2013, doi: 10.1109/TPAMI.2012.159.
Driven by FURP(FoSE Undergraduate Research Placement) Programme.
Multiple Target Tracking by Learning-Based Hierarchical Association of Detection Responses
Publisher:
IEEE
Authors:
Chang Huang; Yuan Li; Ramakant Nevatia
Index Terms - Multiple Target Tracking - Hierarchical Association - Bag Ranking - AdaBoost
Background(Key Point):
The Maximum A Posteriori (MAP) problem of Data Association-based Tracking(DAT)
There are two main components of a DAT approach:
One is a tracklet affinity model that measures the
likelihood of two tracklets belonging to the same target, providing
fundamental evidences for DAT;
The other one is the association optimization framework
that determines which of the tracklets should be linked considering
affinity measurements correspondingly.
The MAP Problem considers initialization, termination and transitions of tracklets, together with the false alarm possibilities of the tracklets, which can be effectively evaluated by the Hungarian algorithm.
The tracklet affinity model, which measures the likelihood of two tracklets belonging to the same target, is a linear combination of automatically learned weak nonparametric models upon various features. which is distinct from most of previous work that relies on heuristic selection of parametric models and manual tuning of their parameters.
Methodology:
Propose a DAT approach based on the hierachical association framework and a novel bag-ranking approach for learning tracklet affinity models which is derived from.
Key Findings:
The systematic experiments on the standard CAVIAR dataset and the
highly challenging TRECVID08 dataset show that this
approach achieves significant improvement in tracking accuracy over
previous state-of-the-art work.
The main difficulty in applying this tracking framework to a real-time tracking system is that it has to postpone “unsafe” associations until enough evidence is collected to resolve the ambiguity, resulting in inevitable tracking lag for some difficult targets.