Virtual Gym Trainer Using CNN

Abstract

This paper presents a CNN-powered virtual gym trainer that utilizes computer vision and machine learning to monitor physical activities and estimate walking distance without wearable sensors. The system uses real-time video input, extracts key points via pose estimation models, classifies activities using a CNN architecture, and calculates physical distance covered using Euclidean geometry. This novel integration of CNN-based classification with motion tracking facilitates contactless health monitoring, catering to home workouts, rehabilitation, and elderly care. Experimental evaluations demonstrate the feasibility and efficiency of the proposed system using a small custom dataset of joint coordinates. The results validate that accurate motion analysis and distance tracking can be achieved with lightweight AI models.