A Graph Transduction Game for Multi-target Tracking
share
Semi-supervised learning is a popular class of techniques to learn from labelled and unlabelled data, especially methods based on graph transduction are widely used. This papers proposes an application of a recently proposed approach of graph transduction that exploits game theoretic notions, to the problem of multiple people tracking. Within the proposed framework, targets are considered as players of a multi-player non-cooperative game. The equilibria of the game is considered as a consistent labelling solution and thus a estimation of the target association in the sequence of frames. People patches are extracted from the video frames using a HOG based detector and their similarity is modelled using distances among their covariance matrices. The solution we propose is effective on video surveillance datasets are achieves satisfactory results. The experiments show the robustness of the method even with a heavy unbalance between the number of labelled and unlabelled input patches.
READ FULL TEXT