Study Of Some Key Technologies Of Three-Dimensional Motion Measurement Based On Stereo Vision

With the rapid development of science and technology, the motion detection equipment used in the fields of the industry, agriculture, traffic, medical treatment and aviation demands higher and higher detection precision, more and more flexible detection means. To meet these demands, the technology and equipments for three-dimensional motion measurement have developed rapidly in the past few years. These systems are widely used in machine building, avigation, electronic industry, medical treatment, vehicle monitor, sport motion analysis, virtual reality as well as numerous other motion measurement tasks. As a novel motion measuring technology, the vision-based approach liberates conventional measuring methods from the astriction of contact, and provides new approach for many relevant problems in the motion measuring fields.The subject"Study of some key technologies of three-dimensional motion measurement based on stereo vision"is to study a new stereo vision three-dimensional motion measurement method based on the feature marker rod from the present researches on three-dimensional vision motion measurement. For this system, we must solve some key problems, such as rigid object three-dimensional motion modeling and computation, edge detection and extraction, corresponding feature matching for stereo sequence images and circle's center extraction, etc.Based on the process of the vision motion measurement and the characteristic of the stereo vision motion measurement system, the feature marker rod, composed of five strong retro-respective spheres, has been designed as the discrete feature. Combination with optimum design of the feature marker rod structure, the imaging quality of the feature points can be dramatically improved, the anti-jamming ability of noise and background can be heightened, the method of feature points identification and matching can be simplified, the feature detection, feature extraction and feature matching accuracy can be ensured.The key step is to set up the motion model for three-dimensional rigid motion object in the process of the motion measurement. The paper designes a kind of motion model and motion equation relative to the rigid motion object's center of rotation according to the reasonable assumption from the rigid object's motion space continuity and time smoothness. It can be applied to measure the center of rotation and correlative motion parameters of the general rigid 3D motion objects while the object's special motion rule and figure are not known in advance. The least of square algorithm and LMedS algorithm are applied to decrease the noise affection in the motion measurement result, it ensures the motion detection is accurate. Finally, the motion algorithm is validated by the simulation and real experiment, the uncertainty of this measurement system is figured out by test.From the difference between the dynamic sequence images and static images, by using an illumination-independent improved wavelet scale multiplication edge detection and fuzzy enhancement algorithm, the feature edges can be accurately orientated in uneven illumination motion sequence images. The edge detection experiments for synthetic and real non-uniform weak illumination images were conducted, the edge detection method's performance was analyzed and compared with other edge detection methods quantitatively and qualitatively. The experiment result indicates it is accurate.Based on the motion measurement and imaging system's character, the paper presents a feature matching algorithm based on a five points cross ratio invariant principle and the corresponding marker rod configuration. The algorithm is proposed from the fact that the cross ratio value is an invariant in perspective projection: the value is unchanged, regardless of the viewpoint, and the corresponding matching can be completed between the left and right cameras, and between the different frame images for the same camera. Finally, the experiment result indicates the matching method is correct and ascendant.According to the characters of the sphere object's image and the requirement of the motion measurement system based on long sequence stereo images, an improved circle's center extraction method is proposed based on the object geometry character's Hough transform algorithm, the computation cost is greatly reduced because of the sufficient consideration about the of the circle object. So the motion measurement's time is greatly decreased. The algorithm is ascendant and correct by the simulation and real experiments.