Due to the expanding application field of gaze estimation, gaze estimation has become an important research topic in human-computer interaction and computer vision. Most of the current mainstream methods for gaze estimation during HCI are based on a sing-camera and single-screen system. In order to address the issues of gaze estimation for multi-camera multi-screen systems, this paper explores the use of a deep learning model, the residual neural network (ResNet50) plus an attention mechanism, to estimate the operator's point of gaze (PoG) on the screen in the flight simulator. First, the front face image is obtained from the three images by preprocessing, and then ResNet is used to regress the PoG. The results show that the inclusion of the attention mechanism enhances the network effect, in which the smallest prediction error is found for on-screen PoG using ResNet50+ECANet compared to SENet and CBAM.