Mexican researchers are developing a neural network that will allow small robots to learn how to distinguish humans from other objects, a capability that could be particularly useful in rescue operations.
Created by University of Guadalajara researcher Nancy Guadalupe Arana Daniel, the system uses a flashlight and a stereoscopic HD camera mounted on a robot to acquire images of the scene. A subsequent application of a sophisticated algorithm enables the machine to recognise silhouettes and distinguish between people and debris.
First, the robot obtains silhouettes and applies a descriptor system to determine visual characteristics (3D points) to segment the object and decide whether it could be a human silhouette. These silhouettes will serve as descriptors to train a neural network called CSVM, developed by Arana Daniel, to recognise patterns autonomously.
"Pattern recognition allows the descriptors to automatically distinguish objects containing information about the features that represent a human figure; this involves developing algorithms to solve problems of the descriptor and assign features to an object," Daniel explained.
In the final step, the software transforms captured images into numerical values representing the shape, colour and density. When merged, these figures give rise to a new image, which passes through a filter to make the final decision.
The researchers would like to make the system capable of independent learning to recognise humans based on its previous experience, mimicking the learning processes of intelligent beings.
The robot used in the research was equipped with a friction crawler-based drive system, similar to that used in war tanks, which can handle all types of terrain.
The computer responsible for the smart operations could be either integrated into the robot, or alternatively, running on a separate computer connected with the robot through a data interface.