A Method For Indoor Robots Visual Perception Based On Multi-feature Fusion

In current life, robot technology has been applied more and more, content-based robot image retrieval has become an active research topic in robotics, computer science and their related disciplines. On the basis of existing content-based robot image retrieval techniques, a method for indoor robots visual perception based on multi-feature fusion is researched.Muratec RAC87A mobile manipulator and personal computer are the hardware of experimental system, Visual C++6.0 and OpenCV are the software of experimental system. Firstly, the robot image retrieval system structure for experimental system is proposed. The features of color histogram, Gabor wavelet texture feature and SIFT are chosen for matching, which are measured by Euclidean and Manhattan vector distance functions. Average accuracy, average recall ratio and average time are chosen as evaluation parameters for multi-feature fusion matching model. For the disadvantages of traditional multi-feature fusion method, the hierarchical match model is proposed. The model is easy to calculate data and can more effectively avoid the shortcomings of various features, accurately and efficiently identify target images.In the experiments,100 real-time collected images of indoor robot and 1000 images of database include the same 50 kinds of indoor objects. The hierarchical match model is used to match. Experimental results show that the proposed model not only achieves a high matching average accuracy, but also meets the real-time requirement of robots. It is suitable to use and promote the hierarchical match model in some application areas.Finally, the main research results are summarized, and the future research directions of the mobile manipulator, which is the experimental platform in this thesis is discussed.