A temporal feature based image registration algorithm is proposed, which achieves effective registration of visible light and infrared non fixed dynamic target images through the analysis and extraction of temporal grayscale features. The focus was on analyzing the anti noise ability of the algorithm. When noise has a significant impact on registration accuracy, a preprocessing filtering method was designed to reduce the impact of noise on temporal feature extraction,thereby reducing the algorithm's mismatch rate. Using candle flames as the detection target, experiments have shown that the algorithm has good noise resistance at an image signal-to-noise ratio of 10 dB, and can achieve visible and infrared image registration of non fixed targets. When the signal-to-noise ratio of the target image is equal to 6dB, the algorithm cannot complete image registration. In this case, the preprocessing filtering method is used to reduce the mismatch rate of the algorithm and successfully achieve visible and infrared image registration of non fixed targets.The key parameter of the preprocessing method calculated in the experiment is 0. 25, which has the best effect.