Research On The Key Technology Of Visual Searching Based On Indoor Mobile Service Robot

With the extension of the mechanism of human perception and the development of computer technology, the advanced control of intelligent robot has become a hot spot in academic.As the rapid development of computer science, image process and pattern recognition, people are paying more and more attention on the robot vision's research and application. And the research has obtained some outstanding results. To make the robot have the function of looking is one of the main questions in the robot vision research. And the process of recognize the environment by robot vision is not simple. Therefore, it has a vital significance and theoretical value to the research of vision recognition system with wide range applicability and high recognition efficiency.This thesis uses the Koala robot and its control system to establish a vision search system on the robot, and introduces the visual attention selection system to improve the searching efficiency. The proposed system consists of three parts:target modeling study, the target areas pre-select algorithm based on visual attention system, and the object recognition based on the matching. The main contents of this thesis are as follows:At first, proposed an image segmentation method based on the SLIC superpixel and CRF conditional random field. Compared with most of the existing methods, the proposed model is more precise and robust and is capable of dealing with uneven illumination, complex background and local textures. With the proposed segmentation algorithm, proto-objects will be divided into several parts.Secondly, the mechanism and the models of visual attention selection system are introduced. A target areas pre-select algorithm is presented based on the VOCUS visual attention selection system. And then, training the object model based on the segmentation method proposed to get the salient regions related to the target.Thirdly, the object which selected by the visual attention system in the salient regions is recognized using the image matching algorithm. According to the different nature of the appearance features, different objects are recognized respectively by SIFT feature point matching or color histogram matching, which is realized by an adaptive selection algorithm.At last, the hardware system of visual searching robot on the Koala robot is established. The serial port communication grogram design between the upper machine and the lower machine is finished. The hardware system debug of the robot is completely implemented. Based on the test of the framework of target searching proposed in this thesis, it can realize the task to searching objects in complete environments by the mobile robot.The proposed visual robot system, with selection mechanism of visual attention, is able to realize the functions to study and search the different types of objects. The system proposed in this thesis has the characteristics of wide applicability and high efficiency.