Content area
Abstract
A facial thermography system was developed to assess the feasibility of detecting mental workload through the use of facial temperature analysis. The system consisted of a thermal infrared camera, a motion tracker and a neural network based classifier. To evaluate the approach participants performed at three different levels or workloads (low, medium, and high) and seven regions of interest were analyzed. Mental workload was correctly classified 81.0% of the time using a network trained with all participants data. When trained using only a single participants data a classification rate of 98.9% was achieved.
I looked at what patterns existed as mental workload increased and found that no easily recognizable trends were present. This is contrary to past studies where cooling of the nose was associated with learning rates and a warming of the periorbital region was associated with stress. Regardless, a complex pattern does exist for the artificial neural network to learn.
