A robust gan-based model for low-light image enhancement and human detection on mobile devices

Detecting people in low-light environments makes it difficult to recognize them with the naked eye, and this can be affected by weather conditions or the time of day. This reduces the reliability of people detection models, as they attempt to detect a shape that is altered by darkness. When the lighting decreases, the image changes, in addition to the noise generated by the lack of clarity. To address this problem, a Generative Adversarial Network (GAN) model is proposed to reconstruct a dark image by eliminating the details that prevented visualization. This is achieved through a concept called ratio-log, which integrates lighting inputs. The results obtained were 25.15 PSNR and 0.909 SSIM using the LOLv1 dataset, and a different result of 28.04 PSNR and 0.905 SSIM using the surveillance-oriented dataset, both showing good results. Furthermore, testing using the YOLOv12 model revealed certain errors when attempting to identify people in dark environments with noise. Integrating image enhancements showed an improvement from 0.44 to 0.78, revealing individuals previously hidden in dark images. Additionally, the GAN model was integrated into an Android application built with a clean architecture, enabling image enhancement within the mobile device. Finally, a survey of IT professionals yielded moderately positive results, confirming its accessibility for users and for future developers who wish to modify it in new research.

Name thesis:degree_name

Ingeniero de Sistemas

Level thesis:degree_level

Titulo profesional

Discipline thesis:degree_discipline

Ingeniería de Sistemas

Grantor dc:publisher

Universidad de Lima

Year dc:date.issued

2026

Author dc:creator

• Camayo Ramos, Sebastian Enrique

Advisor dc:contributor.advisor

• Escobedo Cárdenas, Edwin Jonathan

Licence dc:rights.uri

• https://purl.org/coar/access_right/c_abf2
• https://creativecommons.org/licenses/by-nc-sa/4.0/

Language dc:language.iso

eng

Handle dc:identifier.uri

https://hdl.handle.net/20.500.12724/24823

OAI identifier oai:identifier

oai:repositorio.ulima.edu.pe:20.500.12724/24823