Thermal imaging is used in medical diagnosis and preventive screening, e.g. breast cancer, cardiovascular disease, and orthopedics. Segmentation algorithms fail to recognize body parts of interest when the temperature difference between the body parts and the background is insufficient. We propose to perform segmentation in two stereoscopically acquired RGB images and to triangulate corresponding points extracted from those images into world coordinates. The thereby acquired world coordinates are projected into the thermal image plane for a more robust segmentation result. Our worked example is the segmentation of human hands. The extension of the thermal setup with two additional RGB cameras improves segmentation in our particular case, but could also make segmentation of other body parts in thermal images more robust. Comparing significant points like fingertips and the junctions between the fingers and the metacarpus, we come up with an average deviation of 1.03 pixel +- 0.82 pixel in x-axis direction and 1.04 pixel +- 0.62 pixel in y-axis direction, roughly corresponding to a mean Euclidean distance of 1.4 mm on the hands.