FitVision: Vision-Based Intelligent Fitness Assistance Using Pose-Guided Movement Analysis

Abstract: With the rise in home-based fitness activities, there is a growing need for systems that can guide fitness activities without the need for a trainer. In this paper, we propose a vision-based fitness assistance system called FitVision. The system includes real-time pose estimation, motion tracking over time, and a conversational interface. The system utilizes MediaPipe BlazePose for keypoint detection from video inputs. Based on keypoint detection, a rule-based approach involving joint angles is employed to interpret various exercises. Postures are also checked through joint angle calculations, while repetitions are tracked through a simple state-based approach. Besides the visual component, there is also a chatbot component known as FitBot, which assists users in their fitness-related queries. Moreover, the interaction becomes even more engaging. Based on our results, we observed that the system performs satisfactorily in identifying the exercises correctly, along with a reasonable repetition count. This shows that the use of computer vision along with a chatbot interface may be beneficial in creating a simple fitness assistant.

Keywords: Computer Vision, Pose Estimation, Fitness Monitoring, Deep Learning, Exercise Analysis