In order to detect dim infrared targets from a mass of high resolution images in wide field of view system rapidly and accurately, a novel target detection method based on a phased strategy for research and multi-feature fusion is proposed in this paper. First of all, a saliency detection algorithm based on frequency-domain is carried out to extract candidate region which may contain targets. Then, invariant corner detection algorithm is adopted in candidate region to determine the presence of suspicious targets. Finally, the real targets can be confirmed by time-space coherence in multi frames. The experiment proves that the proposed method can detect dim infrared targets with small calculating amount, high detection probability and low false alarm rate. It is suitable for Infrared Search and Track system in practical engineering.