People counting plays an important role in many people-centric applications including crowd control, traffic management and smart home energy management. With the advancements in wireless sensing, it is now possible to intelligently sense the presence of people with wireless signals. Yet, a lot of challenges arise when Wi-Fi solutions are used for counting humans due to the uncertainty of the states in the environment. In this paper, we propose a novel 3D-Convolutional Neural Network (3D-CNN) architecture able to extract features from range-Doppler images to count the number of people present in an indoor environment by detecting their movements. We generate the range-Doppler images from a Celeno Wi-Fi pulse Doppler radar that uses the 5 GHz frequency band. To the best of our knowledge, this work is the first to count people based on a Wi-Fi Doppler radar. Our experimental results show that our deep learning model is able to estimate the number of people for up to four with an average accuracy of 89%.