For the detection rate is low and clutter is difficult to suppress under the complex scenes, a novel method for infrared small target detection based on curvature near the ground was proposed. The infrared image was regarded as a three-dimensional surface, and the differences between the target and the background in the shape of the surface was analyzed which was represented by curvature in this paper. Then the Facet model is used to calculate the first-order and second-order directional derivatives of the four directions of the image. Search for the zero-crossing area of the first-order directional derivatives, and construct the directional curvature map by using the second-order directional derivatives of the zero-crossing areas. Finally, the directional curvature map in four directions are fused to obtain the final curvature map and the real target was achieved by an adaptive threshold segmentation directly. The algorithm is validated by infrared sequence images in four different scenes. The experimental results show that the algorithm proposed in this paper is greater than 10 in terms of signal-to-clutter ratio and background suppression factor, which are higher than other algorithms. And the detection rate of 100% can be achieved under the false alarm rate below 6×10－4, which is obviously superior to other algorithms.