Provided is a control method for preventing an accident in a tunnel. In this instance, the control method for preventing an accident in a tunnel includes estimating water amount information flowing into the tunnel based on at least one input information, determining whether it is an emergency situation based on the estimated water amount information, and when the emergency situation is determined, transmitting a warning message to an identification device, and controlling a device for opening and closing an entrance/exit of the tunnel. In this instance, the water amount information flowing into the tunnel is estimated through a deep learning based learning model, and the emergency situation is determined by comparing water level information of the tunnel with a threshold value.