Proposing a novel method to detect weak and small targets in infrared image sequences. Preprocess is imposed on image by use of wavelet multi scale analysis to inhibiter arbitrary background. By utilizing the improved Fisher algorithm, we then threshold the preprocessed images, and separate the targets from the background. Target’s pose estimation is obtained through computing the difference of two near frames that have been threshold. Comparing the experimental results on real IR image sequence to that of automatic threshold on class variance show its effectiveness.