Hybrid Deep Learning Method Quantifies Exposed Skin from Images for Safety Assessment
A novel hybrid approach to computer vision, integrating Mask R-CNN with a color-centric algorithm, was created to measure exposed skin in images for evaluating dermal exposure. This technique utilized 170 images from indoor painting, initially detecting human figures and eliminating background noise before isolating exposed skin. The ratios of exposed skin pixels to body pixels demonstrated roughly 80% concordance with human evaluations. This method presents a scalable solution for deriving semi-quantitative exposure data from images, with future developments aimed at recognizing body parts, detecting personal protective equipment, and conducting video analysis.
Key facts
- Hybrid computer vision method developed for dermal exposure assessment
- Combines Mask R-CNN and color-based algorithm
- Tested on 170 indoor-painting images
- Exposed-skin-to-body pixel ratios showed ~80% agreement with human estimates
- Method removes background interference
- Scalable approach for semi-quantitative exposure data
- Future extensions include body-part recognition, PPE detection, video analysis
- Published on arXiv under Computer Vision and Pattern Recognition
Entities
Institutions
- arXiv