In a public transportation network, it is desirable to distinguish people who are getting into a bus at a bus stop from others in the vicinity. Recent advances in machine-learning-based data processing methods enable an automatic identification of human activities to detect the action of getting into a bus. The aim of this work is to propose a method using energy distribution in both frequency and time estimated by analysing the spectrogram of WiFi RF signals to distinguish a change in energy due to a change of speed of motion during a human activity. The spectrogram is a heat-map visualization of a short-time Fourier transform (STFT) of WiFi RF signals. After detecting a human activity through the spectrogram, we automatically detect the action of getting into a bus by tracking the evolution of the spectrogram. Then we evaluate the performance of the proposed method when it uses 1D and 2D features. The results show that adding frequency of movement as an additional feature increases the true negative classification performance from 38% to 95% for the training set and from 27% to 91% for the test set.