The simulation of pedestrian flow plays an important role in public safety research. However, the enhancement of agent authenticity in simulation environments remains a challenge. Herein, we propose simulating pedestrian crossing behaviour at unsignalized intersections using a deep deterministic policy gradient algorithm to drive a two-dimensional optimal velocity pedestrian model. We constructed a strategy exploration scheme that considered two types of two-dimensional optimal velocity pedestrian models: non-velocity and velocity difference terms. The analysis revealed that the model considering the velocity difference term tended to flexibly select relatively safe actions, thus its action-selection strategy was considered optimal. Furthermore, this model completely avoided pedestrian collisions and ensured pedestrian safety.