Tracking and Feature Extraction of Easily Deformable Object Using Particle Filter and Adaptive Vector Quantization
Takeshi Nishida, Norikazu Ikoma, Shuichi Kurogi, Tetsuzo Sakamoto · IEEJ Transactions on Electronics Information and Systems · 2011
PF-mCRL method is a rapid and robust information extraction method for non-Gaussian probability distribution by combination of a particle filter (PF) and an adaptive vector quantization algorithm mCRL (modified Competitive Re-initialization Learning). In this research, a novel method for tracking and shape estimation of easily deformable object in dynamic scene by using the PF-mCRL is proposed. Moreover, several feature value extraction methods from output of PF-mCRL useful for the robot handling are proposed. Further, effectiveness of this proposed method is shown by a real image experiments.