Capturing Visual Attention: A Methodical Approach to Gaze Estimation Data Acquisition and Annotation
Yijun Liu, Qinglan Huang, Honglin Ye, Xiaochao Fan, Changjiang Li, Zhihua Qu · 2024
Gaze estimation technology has brought significant benefits to the field of assisted driving, with gaze as an input method enabling interaction with vehicle equipment. However, this interaction is realized through deep neural networks and extensive data training, with precise image data and image annotation providing the prerequisites for training. To this end, we design an image data acquisition method, selecting the necessary types and models of cameras, and develop an experimental platform to carry the required hardware devices. Additionally, we introduce related image annotation methods. Through this method of acquisition and annotation, we are able to collect high resolution visual estimation images with accurate labels, along with the ground true data under corresponding scenes. This data acquisition and annotation process supports the training of deep neural networks and the evaluation of models with high resolution and labeled dataset.