An Improved Low-cost Binocular 3D Reconstruction Algorithm Based on Open AI Platform

Abstract

A simple, low-cost and good reconstruction-model method has always been a hot and difficult problem in the field of 3D reconstruction. A binocular 3D reconstruction algorithm based on image object detection technology is proposed. First, two low-cost web cameras are employed to acquire images. After the camera calibration is completed, FAST algorithm is applied to detect the feature points of the two images. Then, the image object is obtained by open AI platform. At the same time, image object is used to determine points: only the feature points located in the object area of the images are described by SIFT. In the next steps, the matching, calculation and triangulation of the world coordinate system are also completed by using these feature points. Finally, OpenGL is utilized to complete the 3D visualization of the object. The simulation result shows that the improved 3D reconstruction algorithm can effectively eliminate the 'glitch' caused by mismatching of feature points. The improved algorithm can greatly improve the accuracy of feature point matching without increasing the hardware cost. Feature-point matching accuracy and intuitive 3D reconstruction-model are demonstrated to show the feasibility and validity of this approach.