Object Recognition Of Mobile Robot Based On Omnidirectional Vision

Object recognition is an important issue in computer vision and image processing field,which was widely applied in the military and automatic field. Omni-directional vision sensoris more and more used in the mobile robot filed because it offer a rich source of environmentinformation in a wide angle of view.The object recognition of mobile robot based on omni-directional vision and movingobject recognition in the case of stationary background are studied in this paper. The so-calledstationary background is that the camera is stationary and focal length is fixed Recognizingthe mobile robot in the scene is the main task of this paper, using moving object detectionalgorithms detect the moving objects in the scene is basic idea of this paper, recognizing themoving objects on the basis of detecting the moving objects; therefore, the core of this paperis moving object detection and recognition.Firstly, the reaseach of common moving object detection algorithms——the principlesof temporal difference and background subtraction,based on the above basis,using two-framediffenence、three-frame diffenence、five-frame diffenence、three-frame diffenence using sobeledge information and the background prediction method based on HR filter、backgrounddifference method based on Codebook model、background difference method based on singleGauss model、background difference method based on mixed Gauss model achieve detectingmoving objects in the static scene through programming.Against the exist questions of abovetwo kinds of method, a novel moving object detection algorithm combing three-framedifference using sobel edge information with background difference method based on mixedGauss model is presented in this paper, the proposed algorithm not only overcoming theshortage of singly using temporal difference and background subtraction method but alsocombining the advantages of both method, experimental results vertify the effectiveness andsuperiority of designed moving object detection algorithm.Secondly, image feature extraction method is also one of important reaseaechs in thispaper.Against the characteristic of omni-directional vision image, this paper deeply researchcolor feature, shape feature, Hu moment invariant and SIFT feature which possess scaling,rotation, zooming invariance, above features are prepared for the back of object recognition.Object recognition method based on moving object detection is reaseached.Usingmoving objection algorithm can obtain foreground image, extracting the feature of movingobjects in the foreground image,object recognition method based on color,shape, Hu momentinvariant and SIFT feature matching are designed in this paper, experimental results vertify above object recognition methods are effectiveness.Finally, on the basis of algorithm research, building a object recognition system ofmobile robot with omnidirectional vision under the VC++6.0platform, using above movingobject detection and recognition algorithm in several different kinds of circumstances,experimental results demonstrate that combing three-frame difference using sobel edgeinformation with background difference method based on mixed Gauss model can accuratelydetect all moving objects in the scene,object recognition method based on SIFT featurematching suffers from relatively small influence of objects detection result,which has broaderappliciability and effectiveness.