Human Activity Analysis In Smart Environments

(AI)驱动的虚拟训练伙伴Remote虚拟健身训练最近出现了大规模的繁荣。这一趋势非常受欢迎，不仅因为越来越多的人从事体育活动，还因为远程训练的可持续性。但是，与教练和训练伙伴的互动仍然严重依赖于直接干预，限制了健身训练的可负担性和定制化。随着计算机视觉和自然语言处理的最新发展，可靠的无标记姿势跟踪和真实语言生成等新工具的出现，可以可靠地跟踪受试者的表现，并提供真实的反馈。该项目旨在创建一个人工智能、自主的虚拟训练伙伴，它能够有效地与用户进行口头和非语言交流，并向用户演示练习。个性化类人代理的目的是增加培训动机，最终提高培训依从性。虚拟培训伙伴将与用户建立初步的融洽关系，在培训前设定目标和动机，在培训期间指导用户，并在培训后提供反馈。调查将集中在什么沟通方法和风格(口头和非口头)最有效地最大化用户的内在和外在的培训动机。将研究包括机器学习和计算智能方法在内的人工智能技术对最佳干预时机和方法的适用性。通过建立集合练习的性能指标，代理将尝试估计疲劳，并通过调整休息时间和练习集合的长度来保持受试者的精神状态。此外，为了实现个性化，我们设想基于与用户的交互来构建用于自主虚拟训练伙伴的自我改进的机制。该项目是与全球、世界领先的工业合作伙伴合作的。

"(AI)-Driven Virtual Training BuddyRemote virtual fitness training has recently seen a massive boom. This trend is verywelcome not only because more people are engaging in physical activity, but also becauseof the sustainability of remote training.However, the interaction with coaches and training buddies still heavily relies on directhuman intervention limiting the affordability and customisability of the fitness training. Withrecent developments in computer vision and natural language processing new tools such asreliable markerless pose tracking and realistic language generation, the performance of thetrainee can be reliably tracked and realistic feedback can be given.The proposed project is about the creation of an artificially intelligent, autonomousvirtual training buddy that has the ability to communicate verbally and non-verbally with theuser effectively as well as to demonstrate exercises to the user. The aim of the personalisedhuman-like agent is to increase training motivation and ultimately training adherence.Thevirtual training buddy will establish initial rapport with the user, set goals and motivation inpre-training, guide the user during the training session, and provide post-training feedback.The investigation will be centred around what communication method and style(verbally and non-verbally) is the most effective to maximise the users' intrinsic and extrinsictraining motivation. The applicability of artificial intelligence techniques including machinelearning and computational intelligence methods for the optimal intervention timing andmethod will be studied. By establishing performance metrics for the set exercises, the agentwill try to estimate fatigue, and by adjusting rest times and exercise set length to keep thetrainee enged. Additionally, to achieve personalisation, we envisage building in mechanismsfor self-improvement of the autonomous virtual training buddy based on interactions with theuser. This project is in collaboration with the global, world-leading industrial partner"