Thermographic imaging accompanied with time-resolved analysis is a promising technique for intraoperative imaging in neurosurgery. However, motion due to breathing and pulse of the patient introduces large inaccuracies to the demarcation of normal and pathological brain tissue. Since movements and physiological processes are both manifested as temperature variations, we employ co-registered visual-light images to unambiguously detect motion. In this article, we propose a feature-based approach which is selected from four best-known methods after thorough performance comparison. Complementing our previous work, we evaluate the performance of our methods by applying a frequency analysis and similarity measurements. Our approach enables an accurate motion correction without affecting physiological temperature shifts. Furthermore, real-time performance of the implementation is enabled by serial acceleration and parallelization methods.