Ultrasonic-based localization of cow teats for robotized milking

This article describes a system for the localization of teats in automated milking systems. The system is based on cheap ultrasonic sensors, which are used for building a (noisy) polar map of distances. The paper focuses first on the processing of the polar map data, in order to determine the position of the potential teats; this task is tackled based on an evidence accumulation approach, in order to keep up with the noise effects. Potential teats are then processed in order to recognize the udder; i.e., the correct configuration of teats. The approach for this task is based on a Bayesian belief network. Experimental activity is presented, which confirms the effectiveness of the system. The contributions of the paper are the methods for handling the noisy output of ultrasonic devices, which could not be found in the literature even though commercial systems, based on ultrasonic localization of the teats, are available.