Although many tasks of lower-level autonomous driving have been solved to a good extent, autonomous driving in the presence of other road users such as pedestrians and cyclists still represents a significant challenge. Therefore, precise modeling of their behavior and integration of related models into simulation environments for autonomous driving is of crucial importance for the development of autonomous vehicles. As part of the thesis, the issue of modeling pedestrian behavior near pedestrian crossings based on recorded data was addressed. At the same time, a review of the literature on the subject was carried out, and the most common indicators used in related analyzes were singled out. For the purposes of modeling and obtaining basic insights, a publicly available set of recorded data containing a significant number of vehicle-pedestrian interactions was selected. Pedestrian behavior modeling with regard to selected recorded data was carried out using the Markov chain method, where the associated matrix of transition probabilities was implemented in the form of a sparse matrix based on a dictionary of keys in order to reduce memory requirements and improve computing efficiency. Validation of the derived model was also carried out using Monte Carlo simulations with respect to recorded data. Finally, the possibility of applying the obtained model for optimal control of the speed of vehicles approaching a crosswalk with pedestrians was considered, and the conclusion as well as the possibility of additional refinement of the model were presented.