Research And Implementation Of Depth Extraction Based On Omni-directional Image And Coded Structured Light

Depth information extraction is an important research in the computer vision field. With the rapid development of omnidirectional imaging technology, depth information extraction based on catadioptric omnidirectional image has become a new research hotspot. This technology can be widely used in many ranges such as scene reconstruction, robot navigation, target detection and tracking etc. In this thesis, we combined the catadioptric omnidirectional with the coding structured light. So the active vision measurement system is formed, and the matching problem of passive vision is solved.The panoramic imaging system designed in this thesis is made up of four members, the camera, the hyperbolic mirror, the projectors and three-axis translating and lifting platform. The main works in this thesis included system calibration, the coding and decoding of structured light and the calculation of depth information. Before these works, this thesis describes in detail the theory of omnidirectional vision and structured light systems.The calibration in this thesis included omnidirectional camera calibration and projector calibration. This thesis uses Taylor series to describe the parameters of the omnidirectional camera, and then calculating the coefficients of Taylor series to calibrate the camera. Using this method can obtain parameters of omnidirectional camera accurately with no priori knowledge of the reflect mirror. Projector is different form cameras. Because it is not a imaging device. So it is difficult to obtain the relationship between the target point on the projected image and the corresponding point on the camera plane. After researching the projector calibration algorithms proposed by previous researchers, this thesis proposed a new calibration method in this paper. Using this method can convert the projector parameter calculation to solving the equation of the ray emitted from the projector.In the part of coding and decoding of structured light patterns, this thesis inspired by the check board pattern and designed a four-way hourglass shape primitives structured light pattern. This pattern is easy to extract feature points, and has the advantages of high stability and strong anti-interference. This thesis also has done intensive studies in generating pseudo-random array of the structured light.In the part of depth calculation, this thesis uses the results of panoramic camera and projector calibration and obtain the 3D depth information by solving the ray intersection.After obtaining the depth information, this thesis reconstructs the target in four ways. Verified by the experiment, the method in this thesis can ensure a large field of view and high accuracy of extracting depth information at the same time. This thesis proposes an accurate projector calibration method by analyzing of the error, therefore the accuracy of depth measurement is further improved.