Real-Time Recognition and Tracking in Urban Spaces Through Deep Learning: A Case Study

Real-time object detection in urban environments is critical for security, transportation, and surveillance applications. This work presents an approach based on the You Only Look Once model for real-time object detection in urban scenarios. The methodology employed includes the collection and annot...

Full description

Saved in:
Bibliographic Details
Published in:IEEE access 2024, Vol.12, p.95599-95612
Main Authors: Villegas, William Eduardo, Sanchez-Viteri, Santiago, Lujan-Mora, Sergio
Format: Article
Language:English
Subjects:
academic retention
Annotations
Data collection
Data models
Education
Electronic learning
Ethics
Monitoring
Object recognition
online courses
Performance evaluation
Public safety
Real time
Real-time systems
Safety management
Security
Student participation
Surveillance
Traffic surveillance
Urban areas
Urban environments
YOLO
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Real-time object detection in urban environments is critical for security, transportation, and surveillance applications. This work presents an approach based on the You Only Look Once model for real-time object detection in urban scenarios. The methodology employed includes the collection and annotation of a diverse dataset, as well as the implementation of an intuitive user interface for real-time monitoring. A detailed method is designed in stages, including semi-supervised annotation techniques and data collection strategies in various urban and lighting conditions. The model was evaluated in urban environments, highlighting its ability to handle variations in the density of objects and unpredictable urban events. The results demonstrate that the proposed model achieves a precision rate of 90% and an average processing time per frame of 16 ms, which is suitable for real-time applications. Furthermore, this implementation can handle multiple objects simultaneously and offers robust responses to rapid environmental changes. We demonstrate real-time precision and efficiency improvement by comparing our model with other widely used approaches, such as Faster R-CNN, SSD, and EfficientDet. Additional metrics such as recall, F1 score, and Intersection over Union, essential for a holistic model performance evaluation, are also discussed. This work contributes to research in object detection in urban environments and offers a practical and ethical solution for real-time security and surveillance. Potential applications of our approach range from traffic monitoring to public safety and event management in urban environments. Ethical considerations are addressed, including privacy protection and bias mitigation, which are critical in surveillance technology.
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2024.3426295