Tableware Detection And Recognition Based On Multi-view Learning

With the decline of the cost of service robot hardware and the continuous development of related theories and technologies,different types of service robots have been widely used in various fields.The main task of the service robot in restaurants is to analyze tableware on the tables and then clear them away.In view of this problem,this thesis mainly carries on the research on tableware detection and tableware recognition.The main work is as follows:(1)In this thesis,a tableware detection algorithm based on multi-view learning is proposed.It uses SLIC algorithm to segment initially the whole image into several superpixels,and then takes these superpixels as nodes and models the adjacent relationship as the edges to construct the graph.Using multi-view image features to construct different graphs,our algorithm learns these graphs comprehensively and then obtains the similarities among superpixels.Assuming that the boundary superpixels of images are background,those superpixels which have the least similarities with the boundary are more likely to be the tableware.(2)A classifier is constructed to recognize the tableware in the images.In order to improve the classification accuracy in the recognition process,a multi-view feature fusion algorithm is presented to learn different image features from multiple views and gets discriminative visual features for training the classifier.Compared to single view feature,feature fusion can make these different features be mutully complementary for enhancing the performance.The experimental results also demonstrate that this algorithm can recognize the categories of tableware better.