A gesture recognition system using Localist Attractor Networks for human-robot interaction

In this work, we propose a dynamic gesture recognition system by applying Localist Attractor Networks method. Comparing with existing learning methods for gesture recognition which require a lot of training data from a number of users, our described method only needs a user to demonstrate all patterns of gestural commands a few times before starting a task. The robot can recognize patterns of new gestures from the user only if these patterns belong to the defined patterns. Furthermore, we provide a flexible and easy-to-use human-robot interface, where the proposed dynamic gesture recognition system is applied to control a mobile robot. Experimental results are given to validate the proposed system.