A Physical Model of Human Skin and Its Application for Search and Rescue

For this research we created a human skin reflectance model in the VIS and NIR. We then modeled sensor output for an RGB sensor based on output from the skin reflectance model. The model was also used to create a skin detection algorithm and a skin pigmentation level (skin reflectance at 685nm) estimation algorithm. The average root mean square error across the VIS and NIR between the skin reflectance model and measured data was 2%. The skin reflectance model then allowed us to generate qualitatively accurate responses for an RGB sensor for different biological and lighting conditions. To test the accuracy of the skin detection and skin color estimation algorithms, hyperspectral images of a suburban test scene containing people with various skin colors were collected. The skin detection algorithm had a probability of detection as high as 95% with a probability of false alarm of 0.6%. The skin pigmentation level estimation algorithm had a mean absolute error when compared with data measured by a reflectometer of 2.6% where the reflectance of the individuals at 685nm ranged from 14% to 64%