Takayuki Umeda, Kosuke Sekiyama, and Toshio Fukuda

Abstract:
This paper proposes a cooperative visual object tracking by a multi-robot system, where robust cognitive sharing is essential between robots. Robots identify the object of interest by using various types of information in the image recognition field. However, the most effective type of information for recognizing an object accurately is the difference between the object and its surrounding environment. Therefore we propose two evaluation criteria, called ambiguity and stationarity, in order to select the best information. Although robots attempt to select the best available feature for recognition, it will lead a failure of recognition if the background scene contains very similar features with the object of concern. To solve this problem, we introduce a scheme that robots share the relation between the landmarks and the object of interest where landmark information is generated autonomously. The experimental results show the effectiveness of the proposed multi-robot cognitive sharing.

Keywords: object tracking, multi-robot, feature evaluation, feature selection, autonomous landmark generation