Study On Vision Information Processing And Data Fusion Of Intelligent Robot

With the development of science and technology, the study in robotics comes into being. A kind of machine that can do everything on behalf of human being is expected to exist under the demand of the revolution of society. From 1959, when the first robot was produced, robotic technology has been developed and improved a lot and today it has became a synthecttcal and advanced subject that combines mechanism, electronics, computer, control sensors and signal processing etc.In this paper, we mainly concentrate on two main techniques of intelligent robots: visual technique and data fusion. In the first part, we focus on the vision signal processing in terms of intelligent robots. Generally, three-dimension image is needed in robotic vision systems, but vision sensors can only achieve two-dimension image. To solve this problem, many methods have been studied. Among them, multi-len method and structural Sight casing method are much more practical. But these algorithm were complex. Therefore, to develop a simple and effective method in terms of this problem is very important. In addition, there are some disadvantages of vidicon, such as in visual angle is limited, the visible region is small and so the achievable information is limited cause we are lack of methods to achieve whole and big scale information. Accordingly, in this paper we brought forward to achieve information using single visual sensor and to expand visuable area of visico by panorama sensor. In the second part, we studied data fusion techniques for intelligent robots. In order to solve obstacle detection problem of robots, data overloading and uncertainty of object recognition, we introduced multi-target tracking theory and rough set theory to robotic data fusion techniques and raised corresponding algorithms.The main contribution of this thesis is as following:Study vision processing methods for intelligent robots. By doing inverse computation to get inverse perspective mapping based on vidicon perspective theory. To the end, we can acquire deep information using single vidicon and so simplifiedthe system structure. Starting from the above, we developed an algorithm for partial path design based on single vision sensor. Consequently, moving robot can capture outerspace information in real time by using single visual sensor under neutral condition to lay out pathway.Study image separation and computation based on panorama visual sensors. Aiming at traditional vision sensors has some disadvantages such as visible area is very small etc., we introduced panorama visual vidicon. We also discussed in detail about geometric properties of panorama vision sensors and some problems required to be considered in practical designing. To this end we present an image separation algorithm. What's more, the experiment results show that this method is feasible theoretically and can be utilized into oriental navigation of robot.Study obstacle detection problem of data fusion based on multi-sensors. By introducing multi-target tracking theory and techniques to obstacle detection systems, we can determine the exact position of the obstacle. We discussed data amalgamation problems directly based on achievable data from some sensor without considering the specific structure of individual sensor. We also put forward the corresponding algorithms with respect to standard linear systems, normal linear systems and nonlinear systems.Study multi-sensor information fusion based on rough set theory. We only use the real data acquired by sensors to syncretize multi-sensor information rooted in property reduction, value reduction, and nucleus and incomplete information system etc. According to complete information system and incomplete information system, we show the corresponding amalgamation algorithms, which provide us a very effective method to deal with overloading data of sensors and information amalgamation for incomplete sensor.