| 英文摘要 | Vision measurement is a new concept in computer vision applications. It applies computer vision techniques to the precise measurement of geometrical sizes and locations of spacial objects. With the advantages of non-contact, high accuracy and fast response, vision measurement has a broad application prospect in the non-contact on-line measurement, quality control and motion analysis. Meanwhile, it has been applied to the aerospace, military equipment, biology, medicine, target identification, equipment manufacturing, industrial measurement, automatic control and many other fields. Although the theoretical basis of vision measurement, namely image processing algorithms, is relatively well developed, there are still many problems to be solved in practical applications. For example, some algorithms can achieve measurement results with high precision, but are too complex to be used when real-time performance is needed. In this dissertation, robot vision measurement methods oriented toward embedded systems are explored, aiming at real-time environment perception for robots. The content of this dissertation can be summarized as follows: 1. Based on SIFT (Scale Invariant Feature Transform) keypoints, a binocular stereovision measurement algorithm is proposed and implemented on an embedded image processing card designed by our research group. Firstly, the computation process of the feature vector is simplified using the area-based matching technique to improve the real-time performance of the algorithm. Secondly, a bi-directional searching strategy is adopted to ensure the correctness of keypoint matching results. Thirdly, a quadratic polynomial is used to compensate the measurement results, boosting the measurement precision by overcoming the common problem in stereovision systems where the measurement error increases with the distance. Finally, the measurement system is tested by experiments and applied to a mobile robot to complete the tasks of autonomous parking and obstacle-avoiding. 2. In order to extract information from unstructured scenes, a vision measurement method based on divergence facula structured light is proposed and implemented in the embedded image processing system. After describing the measurement principle and establishing the mathematical model, engineering details are elaborated. The edges of the facula are classified as lines and curves, and then linear fitting or ellipse fitting is used to improve the measurement accuracy. The effect... |
修改评论