| 3D reconstruction of the scene is an important content of the computer vision. As two important methods in computer vision, the stereo vision fits the scene of rich texture information and have a poor quality in weak texture region. On the contrary, structured light fits the scene of weak texture information and have a poor quality in rich texture region. Because of the difference of two systems, how to strengthen the advantages and weaken the disadvantages in a system become a main content of structured light with stereo structure.The paper combines the advantages of stereo and structured light, and creates a structured light with stereo structure system. This system solves the problem of low match ratio in weak texture region and has a good robustness. The followings are the main contributions of this paper:1. The design of two kinds of light patterns. The scene pattern adopts twice-coding and six different values in RGB channels. The heavy stripes adopt De Bruijn sequence. Each heavy stripe does not share the same channel and divides into two different thin stripes. All the process above increase the resolution and match radio. The color calibration pattern adopts three different RGB channels and shares the same value which adopts ladder type arrangement of the scene pattern. This linear mapping model simplifies the model for solving the cross-talk problem proposed by D.Caspi.2. Twice-decoding. This algorithm is designed based on twice-coding pattern and has different cases of distribution of feature points. Decode the heavy stripes firstly and turn to thin stripes. The coding words of feature points are determined by color recognition in RGB space based on the priori knowledge of color calibration.3. Denoising algorithm based on special pattern and coding words check algorithm. The denoising algorithm aims at the noise points in the stripes of interested area and removes these noise points perfectly. The coding word check algorithm aims at improving the robustness and revises the wrong coding words of twice-decoding algorithm. The idea of this algorithm extracts the most same part coding words sequence by comparing with the coding words of scene pattern, then, diffuses to both sides until all the edge points having whole coding words.4. System construct of structured light with stereo structure. The system is divided into two parts:stereo vision and structured light with stereo structure. All the algorithm can be switched each other. This system combines of gathering module, calibration module, matching module and reconstruction module.Finally, the3D plaster is reconstructed by experiments in this paper. Experiment results show that the reconstruction results have higher density of3D points and a better quality comparing with others. |
