Biomimmetic real-time computer vision

Multispectral or hyperspectral vision consists of capturing, processing and representing in an image the information obtained at different wavelengths present in the visible and invisible (to humans) spectrum. This method of image capture allows the extraction of very useful information for certain applications and is widely used in agriculture and in earth observation satellites. However, it has important limitations, such as the large amount of computational effort required, the difficulty of correctly analysing the data and making it works in real time, and also the low resolution of the images provided.

Our research focuses on creating a set of artificial vision algorithms that can manage and process multi-channel information in real time, with the aim of performing detection, segmentation and semantic instantiation tasks in the field of home robotics. It will mean an advance in the capture of information from the environment in mobile robotics, the monitoring of skin, spots or burns or health and quality prediction in plants, bushes or trees.