Diseño e implementación de sistema portable para detección de venas usando termografía de bajo costo.
Loading...
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
Los dispositivos comerciales de asistencia a la venopunción, aunque efectivos, presentan un costo elevado que limita su adopción en el sistema de salud público. Esta Memoria de Título aborda el diseño, implementación y validación de un sistema de detección de venas de ultra bajo costo, basado en imagenología por Infrarrojo Cercano (NIR), con el objetivo de lograr un rendimiento computacional superior a 15 cuadros por segundo (FPS). La metodología se dividió en el desarrollo de hardware y software. El hardware consistió en la modificación de una cámara web comercial (Argom CAM20) mediante la remoción de su filtro de corte IR y el uso de una lámpara de diodo emisor de luz (LED) NIR (630-850 nm) no controlada. El desarrollo del software identificó una limitación clave del hardware: la luz roja visible de la lámpara saturaba el Canal Rojo (R) del sensor. Se demostró empíricamente que el Canal Verde (G) actuaba como un atenuador natural, preservando el contraste venoso sin saturación, por lo que fue seleccionado como la entrada base (Ibase). Se compararon tres pipelines de procesamiento (P1: Ecualización de Histograma Adaptativo Limitado por Contraste, o CLAHE, simple; P2: Filtro Mediana + CLAHE; P3: P2 + Morfología). Los resultados demostraron que el Pipeline P2 (Equilibrado) ofrecía la mejor relación costo-beneficio, ya que el Filtro de Mediana (recomendado por la literatura) añadía una robustez teórica con un costo computacional nulo (94.1 FPS) en comparación con el P1 (94.6 FPS). El Pipeline P3 fue descartado por ser más lento (81.7 FPS) sin mejora visual. Finalmente, el P2 seleccionado se implementó en una validación de video en tiempo real, logrando un rendimiento sostenido de 71.6 FPS, superando masivamente el objetivo. Se concluye que el sistema es técnica y económicamente viable como una prueba de concepto para el sector público.
Commercial venipuncture assistance devices, while effective, have a high cost that limits their adoption in the public health system. This Thesis addresses the design, implementation, and validation of an ultra-low-cost vein detection system based on Near-Infrared (NIR) imaging, aiming to achieve a computational performance greater than 15 frames per second (FPS). The methodology was divided into hardware and software development. The hardware involved modifying a commercial webcam (Argom CAM20) by removing its IR-cut filter and using an uncontrolled NIR Light Emitting Diode (LED) lamp (630-850 nm). The software development identified a key hardware limitation: the lamp’s visible red light saturated the sensor’s Red (R) channel. It was empirically demonstrated that the Green (G) channel acted as a natural attenuator, preserving venous contrast without saturation, and was thus selected as the base input (Ibase). Three processing pipelines were compared (P1: Simple Contrast Limited Adaptive Histogram Equalization, or CLAHE; P2: Median Filter + CLAHE; P3: P2 + Morphology). The results showed that Pipeline P2 (Balanced) offered the best cost-benefit ratio, as the Median Filter (recommended by literature) added theoretical robustness at a null computational cost (94.1 FPS) compared to P1 (94.6 FPS). Pipeline P3 was discarded for being slower (81.7 FPS) with no visual improvement. Finally, the selected P2 pipeline was implemented in a real-time video validation, achieving a sustained performance of 71.6 FPS, massively exceeding the objective. It is concluded that the system is technically and economically viable as a proof of concept for the public sector.
Commercial venipuncture assistance devices, while effective, have a high cost that limits their adoption in the public health system. This Thesis addresses the design, implementation, and validation of an ultra-low-cost vein detection system based on Near-Infrared (NIR) imaging, aiming to achieve a computational performance greater than 15 frames per second (FPS). The methodology was divided into hardware and software development. The hardware involved modifying a commercial webcam (Argom CAM20) by removing its IR-cut filter and using an uncontrolled NIR Light Emitting Diode (LED) lamp (630-850 nm). The software development identified a key hardware limitation: the lamp’s visible red light saturated the sensor’s Red (R) channel. It was empirically demonstrated that the Green (G) channel acted as a natural attenuator, preserving venous contrast without saturation, and was thus selected as the base input (Ibase). Three processing pipelines were compared (P1: Simple Contrast Limited Adaptive Histogram Equalization, or CLAHE; P2: Median Filter + CLAHE; P3: P2 + Morphology). The results showed that Pipeline P2 (Balanced) offered the best cost-benefit ratio, as the Median Filter (recommended by literature) added theoretical robustness at a null computational cost (94.1 FPS) compared to P1 (94.6 FPS). Pipeline P3 was discarded for being slower (81.7 FPS) with no visual improvement. Finally, the selected P2 pipeline was implemented in a real-time video validation, achieving a sustained performance of 71.6 FPS, massively exceeding the objective. It is concluded that the system is technically and economically viable as a proof of concept for the public sector.
Description
Tesis presentada para optar al título de Ingeniero/a Civil en Telecomunicaciones.
Keywords
Diodos, Sensores, Histogramas, Termografía médica