Multiple hypotheses testing (MHT)-based Track-before-Detect (TBD) method is a widely known algorithm to detect infrared dim small target. However, the hypotheses space in this kind of method is so big that the computation cost and storage requirement inevitably increase, resulting hardly realized in hardware. To solve the problem, a parallel two stage multiple hypotheses testing model is proposed. First, a parallel MHT model is designed to save the possible trajectories. By using the proposed MHT, all trajectories can be parallelly processed and the trajectories search process is simplified, greatly reducing computation cost. Secondly, for each pixel on the current image, the trajectories search range is divided into 4 different areas, and two-stage MHT is performed in each area to obtain the accumulation energy. By using two-stage strategy, the trajectories number is reduced exponentially, greatly saving storage space. Simulations experimental results show the superiority of the proposed parallel two stage MHT in terms of detection probability and computational complexity under low signal-to-noise ratio (SNR) condition. It can detect the infrared point target when SNR is 1.5 with a detection probability of 90.6% and a false alarm rate of 0.1%.