Realizing an automatic responsibilities prediction system for road accident using 3D simulation and knowledge systems.

抄録

With the increasing number of motorized vehicles, road accidents are now a big challenge for all countries in the world. Lot of researches in AI, Machine Learning and Deep Learning are conducting every year to find efficient solutions to reduce road accidents through predictions thanks to the usage of data from previously occurred road accidents. After an accident occurred, police have to make investigation to know the circumstances of the incident and determine the responsibilities of each actor. Sometimes, this task can be time-consuming for police and, therefore, support systems are requested. In our research, we focused on a system that can automatically build a 3D simulation (for visualization purpose) of an accident, given as input a manually made accident report. Our simulator, then, automatically generates labeled training data that will be used by the system for image recognition task to predict the responsibilities of each actor in the accident using a custom trained YOLO model. The simulator, also, generates a sketch of the accident to append to the manually made accident report inputted into the system by the user. Our objective is to create a system that can learn and make work easier and quicker for police and improve the traditional and manual way to determine responsibilities when road accidents occur.