Monocular Multi-Perspective 3D Reconstruction for Robotic Ultrasonic Testing in Remanufacturing

Inspection is one of the main processes of remanufacturing. As an important inspection method, contact ultrasonic testing (UT) needs surface scanning on the objects. The transducer should contact and be vertical to the surface. When implementing UT using a robotic arm, path planning on a complex surface is challenging as the parts are deformed from the original CAD model. Computer vision can be used to reconstruct the object surface to plan paths for the surface scanning application. However, the accuracy of reconstruction using structure from motion (SfM) is still critical. The features of images and feature matching directly affect the reliability of the reconstruction of the model. In this study, a method of object reconstruction is established for robotic UT. Only one monocular camera is installed on the end effector of the robotic arm to acquire images from multiple views. After the image acquisition, an improved segmentation method is used to separate the object from the background. This helps to decrease mismatches of feature points and irrelevant feature points. The features are searched in the region of interest using an improved descriptor to increase the number of features in the object area. Subsequently, the poses and positions of the camera are computed using SfM. The 3D model is reconstructed according to the pose information and feature points. The proposed method outperforms the traditional method and also reduces the reconstruction time.