3D reconstruction is an important research field of image processing,computer vision and computer graphics.Binocular vision can obtain three-dimensional information of target objects by simulating the way of human eye with many advantages,such as non-contact,fast speed,high precision and automation.Also Binocular vision can greatly improve the efficiency of industrial production and has already become a research hotspot.Based on the traditional binocular stereoscopic 3D reconstruction system,the method and algorithm are improved and innovated,and the system function is extended according to the specific application.The hardware uses Stereolabs' second-generation ZED binocular camera kit,with the software development kit (ZED SDK) provided by the kit,the efficiency of three-dimensional reconstruction has been greatly optimized by using NVIDIA's universal parallel computing architecture CUDA.Aiming at the noise of stereo matching process and the introduction error of the original view,the interpolation method and the mathematical morphology smoothing method are used to deal with the introduction of the disparity map repair and refinement link.A more reasonable feature point extraction method is used to optimize the depth value calculating link.In the laboratory environment,the performance of the whole system is tested.The results show that the algorithm is stable and efficient,the system has good reconstruction effect and high cost performance.%三维重建是图像处理、计算机视觉、计算机图形学的一个重要研究领域,双目视觉通过模拟人眼处理景物的方式可以获取目标物体的三维信息,具有非接触、速度快、精度高、自动化程度好等诸多优点,可以大大提高工业生产效率,已经成为人们研究的热点;在传统的基于双目立体视觉的三维重建系统基础上进行了方法和算法的改进与创新,并根据具体应用场合拓展了系统功能;硬件选用Stereolabs公司的二代ZED双目立体相机套件,配合该套件提供的软件开发工具包(ZED SDK),使用NVIDIA推出的通用并行计算架构CUDA,很大程度上优化了三维重建的运行效率;针对立体匹配过程噪声及原有视图引入误差等问题,利用插值法和数学形态学平滑方法进行处理,引入了视差图修复与细化环节;采用更为合理的特征点提取方法,优化了深度值计算环节;在实验室环境下,对整体系统进行了性能测试,结果表明,算法稳定高效,系统重建效果好,性价比高.
展开▼
