LAZIER: A Virtual Fitness Coach Based on AI Technology

Ran Bi, Danyang Gao, Xianzhou Zeng, Qiang Zhu

Abstract: Under the influence of the epidemic, online fitness exercise is becoming more and more popular. This is followed by higher requirements for the intelligence and personalization of sports and fitness. Therefore, this paper proposes LAZIER, a virtual fitness coach built through advanced artificial intelligence technology. Without expensive hardware assistance, users can be provided with the three core capabilities of personalized action demonstration, exercise supervision and exercise evaluation through online video teaching, to achieve intelligent and personalized real-time guidance. LAZIER overcomes difficulties such as the diversity of action videos and the scarcity of quality labeled data in the home fitness and sports scene. By designing universal algorithms and few-shot learning, it has realized the core capabilities that virtual coaches must have. Extensive experiments were carried out on the corresponding core technologies to verify the effectiveness.