Point Cloud Segmentation Algorithm Based on Deep Learning and 3D Reconstruction

In order to accurately reflect the surface defects of parts in three-dimensional space, this paper proposes a defect point cloud segmentation algorithm based on deep learning. The system reconstructs the detected object based on binocular vision, and establishes the point image mapping relationship between the color image and the 3D object point. In addition, according to the point image mapping relationship, the image segmentation and point cloud segmentation algorithm are used to segment the defect area, and the corresponding three-dimensional point cloud of the defect area is obtained. This paper uses YOLACT 's deep learning network to intelligently identify defect types and locations to ensure the detection accuracy and real-time performance of the system. In the process of defect detection, the detection image which only contains the defect area is generated according to the segmentation results of the defect image, which not only improves the detection accuracy, but also improves the detection speed.