Object Tracking by Adaptive Feature Extraction (TA-P4)

Author(s) : Bohyung Han (University of Maryland-College Park, USA) Larry Davis (University of Maryland-College Park, USA)

Abstract : Tracking objects in the high-dimensional feature space is not only computationally expensive and but also functionally inef cient. Selecting a low-dimensional discriminative feature set is a critical step to improve the tracker performance. A good feature set for tracking can differ from frame to frame due to the changes in the background against which the tracked object is viewed, and an on-line algorithm to adaptively determine a distinctive feature set would be advantageous. In this paper, multiple heterogeneous features are assembled, and likelihood images are constructed for various subspaces of the combined feature space. Then, the most discriminative feature is extracted by Principal Component Analysis (PCA) based on those likelihood images. This idea is applied to the mean-shift tracking algorithm [1], and we demonstrate its effectiveness through various experiments.

Menu