Abstract: Detection of small, low contrast infrared targets in complex backgrounds requires a significant improvement in the detective ability of photoelectric theodolites. An adaptive Gaussian high-pass filter was proposed based on a description of the dagree of infrared background complexity. The noise of images was reduced by updating a medial filter, and the degree of background complexity was quantitatively described by variance-weighted information entropy. The cutoff frequency of the filter was automatically adjusted according to the degree of background complexity.This technique resulted in the automatic detection of small, low contrast targets in different infrared backgrounds. The new method was verified by experiment and the result shows the new method can effectively detect small, low contrast targets in different infrared backgrounds.