Acquisition and evaluation of depth data from humans, in robotized industrial environments

Industry 4.0 places great importance on collaborative robotics in industrial production and, therefore, on the safe interaction between humans and robots. In previous decades, robots that were part of flexible manufacturing systems remained isolated from human operators through physical enclosures, but current human-robot collaboration activities require a simultaneous concurrent workspace for both actors. Collaborative robotics security systems have used (among others) artificial vision systems based on red, green and blue additive color model images, as proven solutions in detecting humans within the work area of robots. In recent work, the combination of red, green, and blue additive color model imaging with depth imaging systems has demonstrated low sensitivity to lighting changes and a high positional match between distance data and color pixels in both types of acquired images. However, a detailed review resulted in the absence of databases of industrial robotic environments in this type of images. In this paper, an extensive database is delivered in the format already explained that contains people in robotic industrial settings. Furthermore, evaluating the database using machine learning techniques enables computer vision researchers to gain a better understanding of human detection using a Kinect sensor and convolutional neural networks.