The high-cost low-resolution infrared cameras operating in middle infrared spectral ranges are widely used to detect the thermal fields. In this study, a low-cost high-resolution silicon-based sensor camera operating in near infrared spectral ranges is used to perform the observation of the thermal fields on the realistic specimen surface. In near-infrared spectral ranges, a small temperature variation led to a large modification in the sensor illumination, inducing acquired images with over saturation or poor dynamic range of gray levels. To address this problem, an algorithm was used to precisely adjust the exposure time to acquire images with constant gray level whatever the temperature evolution is, and then used in heating experiment of a steel specimen. Results showed that images with constant gray level could be acquired during the experiment. A special radiometric model was used to perform near-infrared thermography. Based on this radiometric model, the thermal fields on steel specimen surface were successfully reconstructed without measuring surface emissivity.