3D object recognition based on confidence LUT of SIFT feature distance

In this paper, a confidence-based matching method for three dimensional (3D) object recognition is proposed. We are developing a remote control system combined with a camera and image recognition system that can recognize specific objects that a user wants to control. Previously, Scale Invariant Feature Transform (SIFT) feature point-based recognition algorithms have been proposed by numerous researchers. However, it is difficult to apply the conventional recognition methods to remote control systems because home appliances tend to have simple shapes, and thus normally produce very few SIFT feature points. To improve the performance under such low feature count situations, a confidence-based 3D feature point matching method is proposed. This method is a modified Best Bin First (BBF) approach that uses a trained confidence look up table (LUT) for decision making. An evaluation of this method is demonstrated on a dataset of 2,432 images.