In view of the difficulties in detecting long-distance target, such as low SNR, relative movement in frames and noise, a preprocessing algorithm using morphological dilation before accumulation and tracking is presented to enhance the target. Wavelet transform is adopted to remove the correlation of 1/f noise in parallel scanned infrared image and keep down the gaussian noise effectively at the same time. Furthermore, the relationship between threshold, detected probability and faulty detected probability of point target in single image is also expatiated. The experimental result indicates that the algorithm can effectively detect the point target with SNR=2.