Recognition and prediction of driver’s whole body posture model
The driver’s whole-body posture at the time of a collision is a key factor in determining the magnitude of injury to the driver. However, current researchs on driver posture models only consider the upper body posture of the driver, and the lower body area which is not perceived by sensors is not studied. This paper investigates the driver’s posture and establishes a 3D posture model of the driver’s whole body through the application of machine vision algorithms and regression model statistics. This study proposes an improved Kinect-OpenPose algorithm for identifying the 3D spatial coordinates of nine keypoints of the driver’s upper body. The posture prediction regression model of four keypoints of the lower body is established by conducting volunteer posture acquisition experiments on the developed simulated driving seat and analyzing the volunteer posture data through using the principal components of the upper body keypoints and the seat parameters. The experiments proved that the error of the regression model in this paper is minor than that of current studies, and the accuracy of the keypoint location and the keypoint connection length of the established driver whole body posture model is high, which provides implications for future studies.