Visual Load Evaluation Model of Multi-view Monitoring Task Operator

Abstract

To explore an effective evaluation method for operators' visual workload in multi-view monitoring tasks, this study conducted a visual workload evaluation experiment consisting of a pre-experiment and a formal experiment. In the pre-experiment, tasks with 1 to 8 visual search areas (View 1 to View 8) were designed to represent different visual workload levels. Combined with the analysis of participants' behavioural performance and the NASA-TLX task workload scale, View 1, View 4 and View 6 were determined to correspond to low, medium and high visual workload tasks, respectively. In the formal experiment, target search tests were carried out on the three types of views, and electroencephalogram (EEG) and eye movement data of 30 participants were collected. Data analysis showed that 7 EEG indicators (including N1 amplitude at Cz/Pz leads within 100–180ms, P2 amplitude at Pz/Oz leads within 180–260ms, θ wave power and θ/β ratio) and 6 eye movement indicators were all sensitive to visual workload changes, with significant differences between low workload and medium/high workload (P<0.05). Based on these 13 indicators, evaluation models were constructed using the particle swarm optimization (PSO) algorithm combined with machine learning algorithms such as SVM and KNN. The results demonstrated that the PSO-KNN model integrating EEG and eye movement features achieved the optimal performance.

Keywords: Multi-view Monitoring, Visual Workload, EEG, Eye-tracking, Machine Learning

DOI: 10.54941/ahfe1007401