Virtual human pose estimation in a fire education system for children with autism spectrum disorders

Children with autism face challenges in areas like language and social skills, which hinder their ability to undergo regular fire training. Fire is one of the most common and dangerous disaster in real life, making it essential to provide children with appropriate prevention and response education. Virtual humans can offer diverse presentation forms and interact with children with autism, thus better stimulating their willingness to participate. To train the fire safety skills of children with autism, this paper proposes the application of highly realistic virtual humans in a fire education system, aiming to improve their fire safety skills. The results show that this approach effectively enhances the fire safety skills of children with autism. To enhance the realism of virtual humans in the fire education system, this paper improves the 3D pose estimation method and proposes a multi-physical factor pose estimation algorithm. By evaluating the Mean Penetration Error (MPE) and the Percentage Not Penetrated (PNP) it was shown that the pose estimation algorithm achieved higher accuracy with only 6.4% foot penetration. We counted the number of movements and the number of movements captured by the system for all participants in the fire training and showed that the system’s motion capture accuracy was over 90%.