An infrared super resolution reconstruction system was proposed to acquire high resolution infrared images. A mathematical model was established according to the procedure of image acquisition. The effect of down-sampling, blurring, motion, and Gussian noise on the infrared system were discussed. Then, a non-degradation feature based sub-pixel motion estimation method was proposed. On the basis of obtained non-degradation, the normalized root of mean square was utilized to estimate the sub-pixel motion between two frames. Furthermore, drawbacks of the conventional total variation factor were analyzed and improved when it was applied in the reconstruction procedure. The region division method was used to divide the image into smooth regions and detail regions, then the new variational factor was able to adaptive to different regions according to their characteristics, and the detail regions could not be over-smoothed. Finally, the experiments on both synthetic and real infrared image sequences were performed by MM(Majorization Minimization). The results indicate that the maximum error of proposed algorithm is 0.09 pixel and the quality of the reconstructed image is better than those of the other algorithms. The proposed algorithm has higher sub-pixel registration accuracy and rich image details,and is able to reconstruct the sequence of low resolution infrared images efficiently.It is suitable for various infrared applications.