HUMAN ACTIVITY RECOGNATION DARI REKAMAN VIDEO PENGAWAS DENGAN METODE YOLO

Abstract

Advances in technology today, so that human work can be helped easily by using machines. One of the jobs that can be time-consuming is supervising. The supervision carried out requires human labor to oversee the events and activities recorded on CCTV cameras. With the development of technology, conducting surveillance is now easier. By utilizing video recordings containing activities, the design made can produce a program that can detect Sitting, Standing, Studying, Raising Hands, and Clapping activities. These detections can then be summarized into an activity logbook along with the time the activity occurred. The system is made using Python and the You Only Look Once (YOLO) method. The program is expected to be able to accurately detect activities with the specified classes. The results show that the YOLO method can find objects and their activities using the internet dataset and IP Camera dataset which produces the highest mean Average Precision (mAP) of 86.85% for the internet dataset and 99.96% for the IP Camera dataset. And the test results on the best model show an accuracy rate of 80.6% for the internet dataset and 98% for the IP Camera dataset.