Proposal of Detection Method of Driving Scenes Based on Supervised Learning and Applicability Evaluation for Simulation Data

Abstract

This paper proposes a driving scene detection method for car driving behavior data. Based on the supervised learning approach, it is possible to generate a scene detection model by collecting data of scenes to be detected in advance and letting it learn. However, since it is difficult to collect the data of the scene to be detected from the huge amount of driving behavior data, we consider the utilization of the driving behavior data obtained by the driving simulator. Naturally, there is a discrepancy between the simulator's driving behavior data and the actual driving behavior data, and scene detection is difficult even if the simulator's data is learned as it is. The purpose of this paper is to obtain the knowledge necessary for devising a method of applying simulator data.