Selection of a Visible-Light vs. Thermal Infrared Sensor in Dynamic Environments Based on Confidence Measures

This paper introduces a confidence measure scheme in a bimodal camera setup for automatically selecting visible-light or a thermal infrared in response to natural environmental changes. The purpose of the setup is to robustly detect people in dynamic outdoor scenarios under very different conditions. For this purpose, two efficient segmentation algorithms, one dedicated to the visible-light spectrum and another one to the thermal infrared spectrum, are implemented. The segmentation algorithms are applied to five different video sequences recorded under very different environmental conditions. The results of the segmentation in both spectra allow one to establish the best-suited confidence interval thresholds and to validate the overall approach. Indeed, the confidence measures take linguistic values LOW, MEDIUM and HIGH, depending on the reliability of the results obtained in visible-light, as well as in thermal infrared video.