Thermal hand gesture recognition
Human-Machine interfaces (HMI) based on hand gesture recognition allow natural, intuitive, and simple communication. Among the different existing technologies, the computer vision stands out for not being intrusive or invasive: the user has not to wear sensors or any additional hardware, providing a great interactivity experience.
In this project, a visual hand-gesture recognition system using very low resolution infrared imagery has been developed. This kind of images has a lot of competitive advantages over the images taken in the visible light spectrum, due to the better signal-interference ratio and independency of the illumination conditions. The main challenges are: the low resolution of the thermal images (80x60) and the reduced hand size inside the image (it only occupies about a 10% of the image). The developed system is based on deep learning algorithms, using Convolutional Neural Networks with a ResNet architecture.