Integration of Color and Shape Features for Household Object Recognition

Intelligent robots such as domestic service robots (DSR), office robots are required to be able to interact with dynamic and complex environments. In order to carry out the tasks given in such environments, the ability to interact with the objects becomes prevalent. In particular, the DSR need to interact with a household object that is normally being lied in arbitrary positions at the home. To accomplish such a challenging task, the robot has to be able to recognize the object. As human does, a visual-based recognition is most common and natural for intelligent robots. To realize such ability the use of visual information captured from a visual sensor is necessary. Thanks to the second version of Microsoft Kinect (Kinect V2), visual information such as color, depth, and near-infrared information can be acquired. In this study, the captured visual information is then processed for object extraction and object recognition. To solve the problems, we propose a method that exploits multiple features such as color and shape features. The proposed method has incorporated the results of each classifier such as k-nearest neighbor (kNN) using a simple probabilistic method to obtain robust recognition results of household objects. To validate the proposed method, we have conducted several experiments. The results reveal that our method can achieve an accuracy of (84.02 +/- 18.85)% for the recognition of household objects with extreme conditions.