Modeling Viewpoint of Forklift Operators Using Context-Based Clustering of Gaze Fixations

This study proposes a method to find optimal viewpoints for Human Machine Interface (HMI) of teleoperation system using human visual pattern. Forklift operation is used as a case study due to its operation complexity consisting of multiple work contexts such as driving and cargo handling. It is challenging to model human viewpoint because there is usually no prior knowledge of the behavior, complicated steps to process behavioral data, and difficulty to represent dynamic behavior throughout the operation. Therefore, a method is proposed to model human viewpoint using setups in a virtual environment that is reconstructed from a real laboratory environment mimicking the warehouse. Gaze points are measured during experiments in virtual environment, and the clustering methods are used to find natural spherical clusters resembling human foveal vision at different work contexts. Viewpoints of several category of forklift operators are derived from the proposed method, and their common viewpoints at each work context is represented using a piece-wise function defining distribution of cluster centroids from the origin of local coordinate system. Spatial analysis of gaze pattern suggests distribution of gaze centroids are generally spatially independent between work contexts. More importantly, the common viewpoint is spatially correlated with the viewpoints of different category of operators. This suggests the proposed model is representative of the general viewpoint of forklift operation.