Research On Fast 3D Imaging System Based On Structured Light Projection Of Binocular Vision

As a measurement technology that can obtain all three-dimensional information of an object, three-dimensional imaging technology plays an important part in many fields such as workpiece testing, industrial designing, reverse engineering, profile modeling, virtual reality, medical diagnosis and so on. With the development of computer technology, three-dimensional imaging technology are getting more attention and showing a greater application prospects in civilian, industrial and military fields.Among all three-dimensional imaging methods, binocular vision has attracted extensive attentions. Binocular vision can imitate human eyes to estimate distance of objects, which works the same way as biologic vision systems. With simple equipment and fast speed, binocular vision also has limits especially when it comes to an object with poor surface textures, making the arithmetic less accurate and harder to get its depth information.The paper begins with mathematical model of binocular vision. After analyzing the relationship between parallax and depth, the necessity and methods of camera calibration are elaborated. Then the process of image correction is presented, followed by two classical binocular matching algorithms and reconstruction methods.Considering the difficulty in applying the two classical binocular matching algorithms into reality, a new method combining structured light projection and binocular vision is proposed. A special stripe structured light is induced into a binocular structured light system and reliable stripe recognitionalgorithm is achieved, which highly enhanced the recognitionprecision. The system precision and anti-interference performance can be improved by special handling of non-extrema. The binocular structured light system is built and the customized software is written, which ensures highly automatic process including calibration, projection, correction, three-dimensional imaging and displaying.At last, system testing is carried out. The accuracy is verified from different angles. Compared with classical matching algorithm, the proposed one presents a less imaging time and a better imaging quality at the same time. A typical object is chosen and repeatable experiments are carried out. Finally, an average error less than 2mm is obtained. Based on the analyses of the influence of system parameters on imaging precision, some parameters are changed accordingly and experimental results are further improved.For the resolution of 700*500, the system takes about 0.5s for 3D imaging, and the imaging accuracy can reach 1 mm.