Adaptive visual tracking on Euclidean space using PCA

In this paper, we present a simple and elegant tracking algorithm that incrementally updates the covariance matrix descriptor using an update mechanism based on Principle Component Analysis on Euclidean subspace. Here, the target window is represented as the covariance matrix descriptors, computed using the features extracted from that window. The covariance matrix is independent of size so it can be compared to any regions without being limited to a constant window size, also it has low dimensionality. We have used the multivariate Hotelling´s T² test to detect the object which is based on Mahalanobis distance. Also, we have incorporated an update mechanism which is based on PCA to increase efficiency of tracking for longer trajectory. This update mechanism also adapts the intrinsic as well as extrinsic variations effectively. The experimental analysis shows the effectiveness of the proposed approach.