Application-driven development of a thermal imaging-based cabin occupant thermal sensation assessment model and its validation

Abstract

The air conditioning (A/C) of cabins allows for customized control, but manual adjustments may distract drivers, as well as result in energy inefficiency. Several existing thermal sensation models require complex inputs, which are challenging to gather whilst driving. To address this issue, this study developed a non-contact thermal sensation model for cabin occupants based on thermal imaging sensor. To collect actual data used for modeling, an outdoor subject experiment was conducted. In this study, initial training was conducted to compare the performance of six algorithms in building the model, with random forests algorithm showing the best performance. Besides, this study employed the recursive feature elimination (RFE) method with cross-validation algorithm for identifying the key features. In the end, the model was retrained using the selected features. The model that incorporated both environmental parameters and facial-temperature features demonstrated the best performance, with an R² of 0.659 on the test set. Eliminating the hard-to-measure windshield surface temperature resulted in a slight reduction in accuracy, yielding an R² of 0.651. To verify the generalizability of the model, this study further conducted independent validation experiments. The selected model, which exhibited a mean absolute error (MAE) of less than 0.4 in thermal sensation units, was proven to be highly applicable. The results can offer new solutions for automatic control of cabin A/C.