Abstract
Infrared head pose estimation (HPE) is considered as the important step in understanding drivers’ intention. Although significant progress has been made in the HPE tasks, previous literature rarely pays attention to the driving environment. In this study, we focus on the similarities of head pose image information and construct the head pose label. The infrared camera is commonly located in the center of the rearview mirror of the car, which determines the angle range for collecting the head pose. According to the angle range values, the head pose distribution label is learned as the two-dimensional Gamma distribution. Given that the camera cannot be mounted directly in front of the line of sight, traditional HPE methods are not applicable. The contributions of adjacent pose images for the central image pose estimation vary. The constructed Gamma distributions are considered as the groundtruth labels and guide the learning process. Experimental results show that the developed model outperforms state-of-the-art approaches and obtains robust performance on several public HPE datasets.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call