In this paper it is presented to resolve the dithering infrared target detection problem under the condition of low SNR. Accumulation of multiframes is often adopted in low SNR infrared target detection, but for dithering low SNR target, this simple accumulation scheme maybe degrade greatly. In this paper, detection algorithm based on dilation before accumulation of multiframes and track before detection is proposed. The dilation operation enables the energy of the same target in different frames to accumulate effectively dispite of the dithering of target, thus we can reach the aim of enhencing the target. The false alarm candidates are removed in the procedure of track before detection. A preprocessing algorithm using wavelet transform is also adopted to remove the correlation of 1/f noise. At the end of this paper, the whole algorithm is simulated with infrared image sequence. A high performance is obtained and the experimental result indicates that the algorithm can effectively detect the target with SNR=2.