<p>Groove geometry measurement and weld seam tracking are essential for ensuring the manufacturing quality of robotic gas metal arc welding. Linear laser vision sensors are often attached to the front of the weld gun, which creates a blind zone when the groove or seam shape is complex. The novelty of this paper centers on the design principles of an omnidirectional laser vision sensor that integrates multiple circular lasers and cameras to achieve 360-degree active vision sensing around the weld gun. The research challenge arises from obtaining a complete view of light stripes, as the weld gun obstructs both laser projection and camera vision. A machine-learning-based algorithm was developed to calibrate the structured light and measure the geometric features of the weld. Validation experiments confirmed the effectiveness of the implemented prototype system, and the maximum relative accuracies for groove width and height measurements were 9.46% and 9.32%, respectively.</p>