Video Surveillance System with a Deep Learning Approach
Abstract
Abstract— The application of in-depth learning methods has been successfully applied in computer vision task with the ability to learn the features of differences in real world images by directly from the original image by passing layer after layer to get the high dimensions image, in this study we applied the YOLO method approach with network adaptation features based on Darknet-53 on a video dataset recorded by the activities of University of Indonesia Prima (UNPRI) students with are conditions of video with different objects as a surveillance system, based on the results of research into object classification produces an overall accuracy of 93%, but for the classification of objects bikes, buses, and cars have the lowest accuracy of 30% for bikes, 54% of cars and buses by 40% so it is necessary to develop methods to improve accuracy.
Moraref