Nowadays, additive manufacturing technologies (AM) suffer from insufficient or lacking methodologies/techniques for quality control. This fact represents a key technological barrier preventing broader industrial adoption of AM, particularly in high-value applications where component failure cannot be accepted. This article presents a real-time melt pool segmentation and monitoring technique applicable to the direct laser metal deposition (LMD) process. An infrared camera with an InSb detector (resolution of 640 × 480, spectral range between 3 and 5 μm) was used. An algorithm, called gravitational superpixels, is presented. This algorithm can group pixels and generate superpixels based on a block generation technique that compares color similarity and temperature in infrared images. Besides, a color similarity correction is applied to reduce uncertainty in segmentation, as well as for eliminating the image background. The task of extracting edges is based on the law of universal gravitation. A quantitative and qualitative algorithm performance analysis, which uses standard metrics, is presented. The analysis demonstrates better versatility than reduction/feature extraction or image segmentation approaches by high-/low-pass filtering. The experimental validation was carried out, extracting and measuring the molten pool geometry and its thermal signature. Then, measures were compared against ground truth and against results obtained by other similar methods. The proposed gravitational superpixel method has higher precision and performance. Our proposal has a significant potential for monitoring industrial AM processes since it requires minimal modifications of commercially available industrial machines.