基于双目视觉的目标完整点云获取方法

Complete target point cloud acquisition method based on binocular vision

快速获取复杂形状目标的完整三维点云,重构目标的三维模型是目前智能检测和机器人识别与定位的关键技术和难点。基于立体视觉,提出一种图像预处理和优化半全局立体匹配过程的方法。该方法设计了图像预处理算法,与半全局立体匹配和点云配准算法配合使用,获取复杂形状目标完整的三维点云。搭建了硬件实验平台和旋转工作台,获取多角度下目标的左右图像对,在此平台上开发了一套基于双目视觉的点云获取系统,并以复杂形状的齿轮为实验对象对该系统进行测试。测试结果表明,该方法生成的物体三维点云效果良好,为目标三维重构提供方法和理论,实用性较强,并且系统运行速度快,算法处理过程可行。

The key technology and difficulty lie in the current intelligent detection and robot recognition and positioning are quick acquisition of the complete 3D point cloud of the complex shape target,and reconstruction of the 3D model of the target.Therefore,a method of image preprocessing and optimization of semi-global stereo matching process is proposed based on the stereoscopic vision.In the method,the image preprocessing algorithm is designed,which is cooperatively used with the semiglobal stereo matching and point cloud registration algorithm,so as to obtain the complete 3D point clouds of complex shaped targets.A hardware experimental platform and a rotating worktable are set up to obtain the left and right image pairs of the target under multiple angles.A point cloud acquisition system based on the binocular vision is developed on this platform.The system was tested taking the complex shaped gear as the experimental object.The test results show that the object generated by using this method has a good 3D point cloud effect,the method can provide methodology and theories for 3D reconstruction of the target,and has a strong practicability,and the system has a fast operation speed and feasible algorithm processing process.