Push Up Counter Based on YOLOV8 Deep Learning

A machine vision based push-up counting algorithm is designed to address the issues of high workload and low efficiency in current physical fitness assessments. This algorithm uses an improved YOLOv8 for human object detection. Ultralight SimplePose predicts the distribution of key points in the human body during push-ups. Then, an SVM classifier is used to train an SVM model, classify several stages of push-ups, and count them. Model training is one of the key steps in the push-up counter. By utilizing machine learning algorithms, the extracted features and corresponding body key points are trained. The test results show that the algorithm can achieve a human key point recognition rate of 0.965, and the accuracy of push-up counting is greater than 99%. It can accurately recognize the human push-up posture in physical training or assessment, and perform counting function.