Pose Detection For A Parallel Mechanism Based On Binocular Vision

Parallel mechanism and parallel robot have been widely focused in the academy according to their advantages of strong rigidity, powerful carrying capacity, stable structure, high precision,low movement inertia etc. Parallel mechanism with few degrees of freedom, represented by three degrees of freedom, has become research focus of robots according to its advantages of simple structure, good dexterity, large work space, easy to control, low cost etc. In parallel mechanism control, the pose information is an important parameter reflecting the kinematic state. Detecting the pose information precisely can effectively avoid the error caused by calculating the kinematic model pose, which would be beneficial to realize high-performance control of parallel mechanism. In the research of pose detection of parallel mechanism, the detection of end pose is still a difficult problem because the detection equipment is expensive, the detection method is complex, and the range of detection is limited and other reasons. Compared with other methods of detection, machine vision has the advantages of non-contact, powerful adaptability, high cost-effective, which is especially suitable for pose detection of parallel mechanism with multi-DOFs and complex movement trajectory.For the detection of end pose of parallel mechanism, the difficult point of the method based on machine vision lie in how to obtain the pose information accurately and quickly using image processing technology. Howere, the machine vision algorithm is complex, and space matching is difficult to precisely achieve owing to external influences such as indoor lighting changes, light reflection of the background, and noise interference, which affect the system detecting speed and measuring precision.For the timeliness and precision problems of the pose detection of parallel mechanism based on machine vision, binocular vision is adopted in this paper to study the problem of pose detection of a new 3-DOF parallel mechanism, and matching problem of pose detection is studied in priority to solve the problem of poor real-time performance of stereo matching algorithm through optimizing algorithm. Thereby the real-time performance and precision of pose detection of the parallel mechanism can be effectively improved.The main research content is as follows:(1) The paper focuses on how to meet the real-timeness requirements of matching algorithm in the detection of a novel 3-DOF parallel mechanism. To solve the problem of poor real-time ability of matching, the key is to reduce the complexity of matching algorithm. As a milestone of matching, SIFT algorithm can not only deal with the matching problem in the conditions of translation, rotation and scale transformation between images, but also remain comparativelystable ability of feature matching. However, for the matching problem of pose detection of parallel mechanism with comparatively high requirement in real-time, the traditional SIFT algorithm complexity, the complexity of the traditional SIFT algorithm is high, and the matching process is too long. In order to solve the problem of poor real-time performance of the matching process, the Harris-SIFT algorithm is adopted in this paper to realize stereo matching with the consideration of the factor that the images of the new 3-DOF driven redundant parallel mechanism are rigid images and there are many corners in these images. The feature points are extracted by a Harris operator, and a SIFT operator is adopted to describe the feature points and match the images. Hence the result of pre-matching has good stability and real-time performance.(2) The paper focuses on how to improve the real-timeness requirements of matching algorithm in detection of a novel 3-DOF parallel mechanism. In the matching process, the mismatching and error matching problems of the system are mainly from error detection,calculation or assumption. For the problem of mismatching and error matching in the matching process, an improved RANSAC algorithm is proposed by selecting feature points in separated grids in the images and pre-detecting to check temporary model to wipe out the mismatching points in Harris-SIFT algorithm. The improved RANSAC algorithm can ensure real-time performance of matching algorithm, and at the same time improve the correct rate of image matching. As a result, this method improves the accuracy of the 3D pose detection of the mechanism.(3) The paper focuses on how to access to the pose information of the novel 3-DOF parallel mechanism. Firstly, this paper uses the pinhole model, and the parameters of camera are derived by zhang camera calibration. Then, some operation such as de-noising, feature point extraction,the Harris-SIFT based stereo matching, purification is carried to the image of parallel mechanism.Finally, the reference frames of moving platform and fixed platform are built according to the characteristic of the novel 3-DOF parallel mechanism, and take the extracted mark points into the camera model to calculate the 3D pose parameter.(4) The experimental hardware platform of the novel 3-DOF parallel mechanism pose detection system is set up based on binocular vision, and the the software system of pose detection is built by development tool Visual Studio combined with Opencv. Based on the experiment platform, the test is carried out to prove the correctness and effectiveness of the mothed.