Search

Physical fitness is one of the most important traits a person could have for health longevity. Conducting regular exercise is fundamental to maintaining physical fitness, but with the caveat of occurring injury if not done properly. Several algorithms exists to automatically monitor and evaluate exercise using the user's pose. However, it is not an easy task to accurately monitor and evaluate exercise poses automatically. Moreover, there are limited number of datasets exists in this area. In our work, we attempt to construct a neural network model that could be used to evaluate exercise poses based on key points extracted from exercise video frames. First, we collected several images consists of different exercise poses. We utilize the the OpenPose library to extract key points from exercise video datasets and LSTM neural network to learn exercise patterns. The result of our experiment has shown that the methods used are quite effective for exercise types of push-up, sit-up, squat, and plank. The neural-network model achieved more than 90% accuracy for the four exercise types. [ABSTRACT FROM AUTHOR]

The pandemic has changed the way people interact with each other in the public setting. As a result, social distancing has been implemented in public society to reduce the virus's spread. Automatically detecting social distancing is paramount in reducing menial manual tasks. There are several methods to detect social distance in public, and one is through a surveillance camera. However, detecting social distance through a camera is not an easy task. Problems, such as lighting, occlusion, and camera resolution, can occur during detection. The research aims to develop a physical distancing detector system that is adjusted to work with Indonesian rules and conditions, especially in Jakarta, using deep learning (i.e., YOLOv4 architecture with the Darknet framework) and the CrowdHuman dataset. The detection is done by reading the source video, detecting the distance between individuals, and determining the crowd of individuals close to each other. In order to accomplish the detection, the training is done with CSPDarknet53 and VGG16 backbone in YOLOv4 and YOLOv4 Tiny architecture using various hyperparameters in the training process. Several explorations are made in the research to find the best combination of architectures and fine-tune them. The research successfully detects crowds at the 16th training, with mAP50 of 71.59% (74.04% AP50) and 16.2 Frame per Second (FPS) displayed on the web. The input size is essential for determining the model's accuracy and speed. The model can be implemented in a web-based application. [ABSTRACT FROM AUTHOR]