Conventional methods in robot vision use the intensity of light reflected or emitted by objects in order to perform object recognition. However, information contained in the polarization of the light can often aid in the determining of surface properties such as roughness, index of refraction, and spatial orientation. Imaging of such surface properties would facilitate image segmentation and classification of objects in military target recognition, environmental monitoring, oceanography, forestry, agriculture, and automated assembly. Physics Innovations Inc. is developing a thermal imaging technique where, in each image pixel, three Stokes parameters are sensed simultaneously and at video frequencies. The Stokes parameters are intensity I, percent of polarization P, and angle of the plane of polarization (phi) . Although infrared, thermal intensity images of terrestrial scenes have low contrast, images of P and (phi) are expected to have high contrast. In this paper the Physics Innovations sensor is described. We also discuss our evolution of the performance of a prototype sensor. Images of I, P, and (phi) from the prototype sensor demonstrate that, for common man-made objects with smooth surfaces, surface orientation can be derived. Surface orientations can be measured in the same image frame as temperature distribution. From our results using the prototype sensor, we conclude that three-dimensional information, in addition to thermal information, can be derived from polarization-sensitive, thermal imaging.