An active model-based algorithm for correspondence and estimation of pose parameters of objects

摘要

A method of actively corresponding a model with an image and estimating pose parameters is proposed. The algorithm used follows hints from human visual recognition processes, widely believed to be efficient and adaptive. It introduces into typical bottom-up information extraction procedures several top-down information processing ones; these latter are usually effective and robust. The combination of such procedures that form our "active model based algorithm" makes the whole process more effective. The system is robust because of the high level influence on the low level image processing. Experimental results show the parameters obtained are precise enough for robot gesture estimation, navigation and obstacle avoidance.