Q-Traffic Accident Monitoring System Using Passenger Head Pose Estimation and Deep Learning

초록

The existing emergency rescue service generally provide information about the accident, such as the location and time of the accident, but it is difficult to take emergency measures due to insufficient information on the patient's conditions. In this paper, we propose a system that detects the accident situation more accurately through the proposed application system that uses the images of the black box and deep learning based accident recognition algorithm, and provides detailed information related to the injuries of the occupants in real time. In this system, the type of accidents is classified using YOLO to accurately grasp the accident situation. If it is judged as an accident, the head posture estimation technique and HSV(Hue Saturation Value) color model are used to determine the level of injury and bleeding of the occupant and then all related information is sent to the remote server of the emergency rescue agency using the application linked to the occupant's mobile phone. Through performance evaluation based on actual accident images, it was shown that the proposed system detects accidents with 95% accuracy, and the patient's condition related to neck breaks and bleeding has an accuracy of 76-95%. Through the system proposed in this paper, we can reduce human casualties through appropriate emergency measures by identifying the accident status and the damage status of the occupants in advance. © 2021, Korean Institute of Communications and Information Sciences. All rights reserved.

키워드