Video-based sharp instruments detection for children safety using YOLOv3

Object detection using deep learning has a very wide range of applications in several areas. Children in human society could use sharp objects such as knives, forks, hammers, and scissors and injure themselves without their parent’s awareness. Sharp object identification software must be coupled wit...

Full description

Saved in:
Bibliographic Details
Main Authors: Hussein, Rawaa Sabah, Ibrahim, Nuha Jameel, Sadiq, Ahmed T.
Format: Conference Proceeding
Language:English
Subjects:
Algorithms
Artificial neural networks
Boxes
Children
Cutlery
Cutting tools
Datasets
Hammers
Image contrast
Kitchen utensils
Knives
Machine learning
Neural networks
Object recognition
Safety
Surveillance systems
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Object detection using deep learning has a very wide range of applications in several areas. Children in human society could use sharp objects such as knives, forks, hammers, and scissors and injure themselves without their parent’s awareness. Sharp object identification software must be coupled with video-based security and surveillance systems to monitor such activities. In order to implement that, this study proposes to utilize the You Only Look Once (YOLOv3) approach for the detection of sharp objects. In the case where contrasted to other object detection algorithms, the YOLOv3 methodology has some advantages. Other approaches, like the Convolutional Neural Network (CNN) and the Fast-Convolutional Neural Network (FCNN), don’t examine an image thoroughly. In contrast, YOLOv3 does so by predicting bounding boxes with a convolutional network and determining class possibilities for those boxes, in addition to detecting the image in a shorter time than the other approaches. This work represents a method for detecting sharp objects in order to keep children safe. The proposed system employs YOLOv3 on the data set to detect several types of dangerous devices to children, such as knives, forks, hammers, and scissors in different places. The image datasets utilized for training and testing the network in this experiment were extracted from a video. These images are colored, and they are captures of children holding non-safety sharp objects. The data set includes 425 images, where 160 images have been specified for the testing. According to the research findings, the proposed system effectively recognizes objects with 85% accuracy.
ISSN:0094-243X
1551-7616
DOI:10.1063/5.0150739