Weld Bead Localization via Spatial-Temporal Hidden Markov Models

Abstract

To perform non-destructive testing (NDT) of weld beads in the wild environments, we proposed an approach by defining cascaded spatial-temporal Hidden Markov Models (HMMs) on a structured light device. Structured light stripes projected on weld beads are firstly extracted with a spatial HMM (S-HMM) defined on the spatial characteristics of stripes. A measurement function is then defined to detect candidate positions of weld beads in video frames according to the extracted stripes. Finally, based on the candidate positions and utilizing continuous characteristics of weld beads, a temporal HMM (T-HMM) is defined to precisely localize and measure weld beads in video sequences. Experiments show that the proposed approach can effectively localize weld beads and then navigate the robot (a moving platform for inspection) in wild environments.