Target Identification And Positioning Research Based On Binocular Stereo Vision

In the recent years,robot technology has made great progress.It plays a real important role in every area around our life,such as military,aerospace high precision special and in industry,agriculture,medical treatment,smart home.And it is gradually developing towards intelligent and automated direction.As a specific form of the robot,mechanical arm greatly simplifies the people's production and life.Making the mechanical arm has visual functions just like human eyes has becoming the research hotspot.6 degree of freedom mechanical arm is chosen as the research object in this paper.The target identification and orientation of mechanical arm is studied using the binocular stereo vision technology.The main work of this paper is as follows:At first,two categories of camera imaging was introduced in this paper,namely the linear and nonlinear model.The target object's four kinds of coordinate system in several different space and the reciprocal transformation between them was introduced in detail.The principle of binocular camera measurement and imaging was presented,and the Open CV calibration method was used to complete the camera calibration.The internal and external parameters of camera was solved.Secondly,in view of the target object matching and recognition,the common method to get images' characteristics SIFT algorithm was introduced in detail.Aiming at SIFT algorithm's insufficient,the local invariant features SURF operator was chosen to stereo matching of target object recognition.Then the matching result was purified using PROSAC algorithms,the target area is affirmed based on homograph matrix,and the transformation between the target area and source image coordinates was deduced.Finally,the target location problem was studied.The target contour was further extracted by introducing SURF algorithm into graph cut algorithm Grab Cut.The Grab Cut algorithm's operation process was also simplified.Aiming at the problem that the center image positioning of target image from two around cameras are different.The target object's contour point center is adjusted by polar constraints and template matching algorithm.Then the object pose estimation is achieved.