In most of Japanese commercial districts, pedestrian count surveys have often been conducted to count the numbers of pedestrians who pass through several streets in order to grasp the pattern of pedestrian flows in their district. While they are useful to compare pedestrian flows at different points of time if conducted regularly, the data obtained from such surveys have never been utilized further. This paper proposes to introduce a new method of the pedestrian count survey so that the data obtained from this method enables one to construct a hidden Markov model to estimate dynamical pedestrian flows on the streets in a commercial district. We also give an actual application of our method to the case of city center retail district at Fukuoka City and show that our method is effective enough to have wide applicability.