Application of color segmentation using online learning for target tracking

Abstract

Color segmentation techniques find extensive applications in visual tracking as the color information provides a robust reference for identifying a target. Color based tracking systems generally use histograms or static models. However, in the real world the changing surrounding conditions must be taken into account. An online learning method of color segmentation has been adapted to ensure better performance even with changing lighting conditions. A neural network, based on fuzzy Adaptive Resonance Theory (ART), is used to develop the color model that is updated with each frame by the pixels classified within the target. The categories formed by the ART network are restricted to ensure fast processing, and the performance of the system is analyzed at different thresholds for association with the color model. Further, a Kalman filter is added into the loop for predicting the target's position in the next frame and a comparison is made to investigate the improvement in performance. © 2008 Springer-Verlag.