主催: 一般社団法人 日本機械学会
会議名: 第29回交通・物流部門大会
開催日: 2020/11/18 - 2020/11/20
Advances in sensing and control technologies have led to rapid improvements in safety devices to avoid hazardous events. However, when the driver's does not operate the vehicle in automatic operation and the system requires sudden operation, the sudden change in vehicle behavior by the system may put passengers in danger. The purpose of this study is to realize the discovery of the driver's hazard prediction in normal driving by deep learning. In this experiment, we used deep learning to categorize the forward images based on the unconscious braking/accelerating operations and the resulting acceleration in a situation that does not lead to a near miss accident, and extracted the features. This can prevent sudden operations by detecting the risk of collision earlier. We have shown that it is possible to discriminate situations that do not lead to near miss accidents by investigating the conditions of classification for the early identification of collision risk from images in front of the vehicle.