In order to enhance the safety and efficiency of pedestrian-vehicle evacuation in underground parking lot fires, a background field model based on cellular automata is constructed using Python. The model incorporates fire field data of the PyroSim simulation to create a dynamic coupling model of fire and pedestrian-vehicle evacuation. A simulation study is conducted in the underground parking lot of a commercial complex in Xi’an. The results show that, compared to pedestrian evacuation, the Required Safety Egress Time (RSET) for pedestrian-vehicle evacuation is extended by 20% to 130%. Safe evacuation is only achievable under specific conditions of pedestrian density and parking space occupancy. The arch effect is observed at the exit when pedestrian density is high, which reduces the evacuation efficiency of both pedestrians and vehicles. The probability of vehicle evacuation failure is approximately 22 times higher than that of pedestrians. Even if pedestrians successfully reach the exit, 15% to 25% still suffer from varying degrees of injury, with the proportion of serious injuries and near-death cases positively correlated with the pedestrian evacuation density. These findings can serve as a reference for optimizing fire protection design in underground parking lots and formulating evacuation plans for both people and vehicles.