Estimating Attention Area of Spectator Using Self-calibration in Sports Fields

Abstract

In sports video solutions, estimating the attention area of spectators is useful for extracting important scenes. The attention area is identified as the area on the court corresponding to the gaze direction based on the spectator's gaze. To do this from cameras installed around the sports field that capture the spectators, camera calibration is required to convert the coordinate system of the camera that shoots the spectators and the coordinate system of the court. However, it is difficult to apply the general camera calibration to the camera capturing spectators because the court is not reflected in the camera. Therefore, in this study, we propose a method to estimate the attention area by self-calibrating the combination of the camera capturing the spectators and the camera capturing the overhead view. Since it is difficult to estimate the gaze accurately from low-resolution images of spectators, we approximate the gaze direction as the head direction and reduce the error by aggregating multiple head directions. Verification using the shooting data of an actual 3x3 basketball game shows that a reasonable attention area map can be obtained based on three camera inputs.