Multimodal social scenario perception model for initial human-robot interaction
Human-robot interaction imposes many challenges and artificial intelligence researchers are demanded to improve scene perception, social navigation and engagement. Great attention is being dedicated to the development of computer vision and multimodal sensing approaches that are focused on the evolution of social robotic systems and the improvement of social model accuracy. Most recent works related to social robotics rely on the engagement process with a focus on maintaining a previously established conversation. This work brings up the study of initial human-robot interaction contexts, proposing a system that is able to analyze a social scenario through the detection and analysis of persons and surrounding features in a scene. RGB and depth frames, as well as audio data, were used in order to achieve better performance in indoor scene monitoring and human behavior analysis.