Research On Object Recognition Based On Mobile Robot Platform

Research on object recognition based on robot platformAs is shown by human vision research, the proportion of information obtained by eyes is75%, in other words, visual information is our most important method to obtain information from outside worlds. Thus, with the rise of artificial intelligence research, computer vision has been unprecedented treasured. How to make robots have the ability of understanding visual information as human beings, to identify different objectives, understanding the meaning of the scene to see; and how to make the robot make some intelligent behaviors under the guidance of visual information is a hot research field of computer vision.Object recognition using a computer roughly divided into the following steps:First feature detection, then describe images using uniform methods, then design a object classifier, finally train the classifier to make it have the ability to recognize new images. The classic object recognition algorithm is Bag-of-words model, part-based model and the discriminative model. In this paper we mainly study how to recognize single object class in an image using Bag-of-words model. We realized Bag-of-words model using C++Language and openCV Library, this can be used in mobile platform, which needs to be real-time. We studied Probabilistic Latent Semantic Analysis classifier, naive Bayes classifier, designed different experiments to verify performance of different classifiers. According to this, we proposed an improved Probabilistic Latent Semantic Analysis classifier, the result showed that our new classifier had a higher recognition accuracy. This paper also combined Bag-of-words model with pattern recognition, using neural network classifier to recognize single object in images. Specially, we studied recognition principle and training method of BP classifier, designed a neural network classifier and realized it using c++language, analysed factors of recognition performance of BP classifier, designed sets of experiments to find changing regulations. At the end of this paper, we did some research on Voyage Ⅱ mobile robot platform, and proposed an available method to apply object recognition algorithm on mobile platform.In this paper, on research of the Bag-of-words model, we can make a image into a vector, on research of pLSA, naive Bayes and BP neural network classifier, we can classify visual vector, so we can recognize object in images. To apply these algorithms on mobile platform, we also realized Bag-of-words model and BP classifier.