People flow counting at the doorway is the focus of current research, which can prevent trampling accidents caused by excessive crowd density in an area. However, most existing systems assume that there is only one direction of people flow at the same entrance, and then count the entering or exiting people through different entrances, respectively, without estimating the direction of people flow. Therefore, to detect the entry and exit direction of the people at the same Entrance/Export, we propose an Empirical Mode Decomposition (EMD) based method to estimate the passing direction. First, we analyze the Doppler frequency characteristics of the people passing through the transceiver link in the doorway scenario. Then, according to the fluctuation characteristic of Doppler frequency and signal amplitude when the people enter or exit, we use EMD to decompose the signal and then obtain several Intrinsic Mode Functions (IMFs) with different frequencies. Finally, the direction of entry or exit is determined by the order of fluctuations of these IMF signals. The experimental results show that the average accuracy of passing direction detection is 96.5%.