Multi-sensor information fusion: Combination of fuzzy systems and evidence theory approaches in color recognition for the NAO humanoid robot

Reda Boukezzoula,Didier Coquin,Thanh-Long Nguyen,Stéphane Perrin

DOI: https://doi.org/10.1016/j.robot.2017.12.002

IF: 3.7

2018-02-01

Robotics and Autonomous Systems

Abstract:This paper investigates the conception of a real-time decision system for colored object recognition using cheap and uncalibrated cameras. The proposed decision system, embedded in a humanoid robot evolving in an uncontrolled environment, is based on the collaboration between a fuzzy system and a data multi-sensory fusion methodology. Based on the ability of fuzzy systems to exhibit any type of behavior mapping, the color recognition problem using a single camera is viewed as a Takagi–Sugeno fuzzy system with Constant Conclusions (TSCC). Indeed, the behavior between the HSV triplet extracted from the image (captured by the robot’s camera) and the recognized colored object is represented by a TSCC. In this context, it is well known that a single sensory system is only capable of supplying partial information and is consequently limited in its ability to interpret complex situations, especially in imprecise and uncertain environments where ambiguities and conflicts among colors are often present. In other words, a decision system based on a single camera cannot reflect real situations and can affect the reliability and the quality of the color recognition. In order to overcome this problem, the integration of more than one camera in the recognition process is proposed. The data provided from these cameras is always affected by some level of impreciseness as well as uncertainty in the measurements. So, the proposed multi-camera fusion strategy based on evidence theory should be able to express such imperfections effectively, to reduce their effects, to exploit the data redundancy, to integrate the reliability of sources and to improve the color recognition accuracy. The feasibility and the validity of the proposed methodology are experimentally validated according to an embedded real-time implementation on the NAO humanoid robot.

robotics,automation & control systems,computer science, artificial intelligence