Publication: Software de reconocimiento de piezas de ajedrez en tiempo real
Loading...
Official URL
Full text at PDC
Publication Date
2022
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
El reconocimiento de las piezas de ajedrez en un tablero físico es un problema de visión que aún no se ha resuelto de manera eficiente. Se han logrado algunos avances a lo largo de los años, pero aún carecen de la precisión o el rendimiento suficientes para usarse a nivel práctico para digitalizar juegos en vivo. Este proceso de digitalización actualmente se realiza a nivel profesional con la ayuda de tableros de ajedrez y piezas especializadas que no están al alcance de la mayoría de los aficionados al ajedrez debido a su costo. Hay un interés renovado en encontrar nuevos algoritmos para este problema, gracias al éxito de las redes neuronales convolucionales. Sin embargo, aún existen varias dificultades para su adopción práctica, especialmente en implementaciones que intentan satisfacer las restricciones de tiempo real de los juegos en vivo con recursos de hardware de presupuesto limitado. En nuestro TFG, hemos tratado de superar algunas de estas limitaciones portando a una aplicación de Android algunos de los algoritmos más prometedores propuestos en este campo. Además, también hemos intentado mejorar el entrenamiento de las redes neuronales diseñando e implementando un dispositivo de bajo coste que simplifica la generación del dataset de entrenamiento.
There cognition of chess pieces within a physical board is a vision problem that has not yet been solved efficiently. Some progress has been made over the years, but they still lack enough precision and or performance to be used at a practical level to digitize live games. This digitization process is currently performed at a professional level with the help of chess boards and specialized pieces that are not available to most chess enthusiasts due to their cost. There is a renewed interest in finding new algorithms for this problem, thanks to the success of convolutional neural networks. However, there are still several difficulties for its practical adoption, especially in implementations that try to satisfy the real-time constraints of live games with budget-limited hardware resources. In our TFG, we have tried to overcome some of these limitations by porting to an android app some of the most promising algorithms proposed in this field. In addition, we have also tried to improve the training of the neuronal networks by designing and implementing a low-cost device that simplifies the generation of the training dataset.
There cognition of chess pieces within a physical board is a vision problem that has not yet been solved efficiently. Some progress has been made over the years, but they still lack enough precision and or performance to be used at a practical level to digitize live games. This digitization process is currently performed at a professional level with the help of chess boards and specialized pieces that are not available to most chess enthusiasts due to their cost. There is a renewed interest in finding new algorithms for this problem, thanks to the success of convolutional neural networks. However, there are still several difficulties for its practical adoption, especially in implementations that try to satisfy the real-time constraints of live games with budget-limited hardware resources. In our TFG, we have tried to overcome some of these limitations by porting to an android app some of the most promising algorithms proposed in this field. In addition, we have also tried to improve the training of the neuronal networks by designing and implementing a low-cost device that simplifies the generation of the training dataset.
Description
Trabajo Fin de Grado en Ingeniería Informática, Facultad de Informética UCM, Departamento de Arquitectura de Computadores y Automática, Curso 2021/2022.