In order to detect infrared dim targets in low Signal-to-noise Ratio (SNR) background,an adaptive Track-before-detect (TBD) algorithm based on Directional Median Value Filtering (DMVF) and Layered Dynamic Programming (LDP) was proposed. The structure clutter was suppressed using four adaptive Weighted Directional Median Values (WDMV) on the ground of its correlation in the local neighbor. For the out-threshold potential targets,the LDP was implemented for multiframe detection through the continuity and smooth of the trajectories in order to reduce the computation complexity. As to the matching the trajectories detected in every layer,a decision was made. The simulated experiments show that the SNR is effectively improved by DMVF,and LDP reduces the false detection probability and greatly improves the operation speed.