An Object Tracking-Based Poses Vision Measurement System For Parallel Robot

In order to acquire the control parameters of a robot control system,the position and orientation parameters are usually used to get the inverse solutions the dynamic-geometry formula.The MDOF parallel robot has complicate trails.Generally speaking,we detect these trails in tradition method with the help of sensors such as accelerate sensor, distance sensor,angle sensor,and then the signals we just get are transferred and sent to the system control circuit to control the robot.In recent years,the vision detecting method,because of its non-touch,intelligent,and its fast detecting speed,advantage of measuring multi-degree motion,have been used in many research areas.In this paper,a software structure is presented,which use vision based detecting method to grab pictures of the object in the camera view in real-time,then a kind of single camera detecting measurement is used to acquire the parameter information of the robot motions.In this paper,the system structure based on vision measurement is also presented. Under this structure,a single camera is applied to grab pictures of a 2-DOF redundant actuation parallel mechanism with 3 inputs.Then a vision based object recognition method is used to identify the object location in each of the pictures,and resize pictures as only including the location areas or other areas adjacent to the object location.At last of the processing,we use a kind of affine invariance method to make sure the real position and orientation of the robot.The system it is composed of image acquisition and transfer,camera calibration,affine-invariance,reconstruct of points and position and orientation measurement.Main steps are as follows.First,a simple linear calibration method based on pin-hole model is used to calibrate the single digital camera of the vision position and orientation measurement system for parallel robot.Second,an object recognition processing is built,which uses Adaboost theory to train Haar features abstracted from the object needed for next processing.It performs reliable matching between different views of objects,when the images have change in scale,rotation,shift and change in illumination.It is especially adapt to measurement for multiple DOF parallel robot and complex motions in space.Finally,the parameter of pose can be calculated by matrixes of translation and rotation of 3D,the matrix just mentioned can be linearly calculated by using some kind of affine-invariance.In this article,a kind of trapezium affine- invariance is presented to get the matrix.All the process is programmed in c with VC++6.0.The results of simulation experiments prove the veracity and validity of all algorithms in this thesis.