Detección de personas en tiempo real mediante algoritmos de Deep Learning
Loading...
Official URL
Full text at PDC
Publication date
2021
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
En muchas escenas existe la necesidad de contar el número de personas que acceden a un determinado espacio para limitar y controlar su flujo, especialmente en los tiempos de pandemia. Por lo tanto, en el presente proyecto se presenta una posible solución para contar las personas de forma automática en tiempo real, utilizando la tecnología del aprendizaje profundo. La solución, enmarcada bajo el paradigma de Internet de las Cosas (Internet of Things, IoT) consta principalmente de dos partes, nodo IoT y servidor. El nodo IoT se encarga de capturar las imágenes mediante un módulo de cámara y las envía al servidor. El servidor es un dispositivo con altas capacidades computacionales, que se encarga de los procesamientos de las imágenes, detectando las personas que aparecen en dichas imágenes mediante el uso de algoritmos de aprendizaje profundo. En el presente proyecto se han utilizado dos algoritmos, Mask R-CNN y YOLACT, con tal propósito, que permiten analizar los diferentes resultados. Conjuntamente con ellos, se ha utilizado el algoritmo de DeepSORT para realizar el seguimiento de objetos, asignando un ID a las personas detectadas. Finalmente, con las coordenadas obtenidas de las personas, se determina su sentido de movimiento y se alerta cuando el número de personas supera el límite establecido en el espacio objeto de monitorización. Todos estos métodos y funcionalidades se han integrado convenientemente hasta lograr una solución conceptual IoT, que ha permitido comparar y evaluar las distintas estrategias integradas.
In many scenes, there is a need to determine the number of people to limit and control their flow, especially in times of pandemic. Therefore, this project presents a possible solution to count people automatically in real time, using deep learning technology. The solution, consists, mainly of two parts, IoT node and server. The IoT node is responsible for capturing images using a camera module and sends them to the server. The server is a high computational device that is in charge of image processing, detecting the people in the images using deep learning algorithms. In the project, two algorithms, Mask R-CNN and YOLACT, have been used to analyse different results. Then, the DeepSORT algorithm has been used to track objects, assigning an ID to the detected people. Finally, with the coordinates obtained from the people, their direction of movement is determined and an alert is given when the number of people exceeds the limit. All these methods and functions are conveniently integrated to achieve a conceptual solution IoT, which has allowed to compare and evaluate the different integrated strategies
In many scenes, there is a need to determine the number of people to limit and control their flow, especially in times of pandemic. Therefore, this project presents a possible solution to count people automatically in real time, using deep learning technology. The solution, consists, mainly of two parts, IoT node and server. The IoT node is responsible for capturing images using a camera module and sends them to the server. The server is a high computational device that is in charge of image processing, detecting the people in the images using deep learning algorithms. In the project, two algorithms, Mask R-CNN and YOLACT, have been used to analyse different results. Then, the DeepSORT algorithm has been used to track objects, assigning an ID to the detected people. Finally, with the coordinates obtained from the people, their direction of movement is determined and an alert is given when the number of people exceeds the limit. All these methods and functions are conveniently integrated to achieve a conceptual solution IoT, which has allowed to compare and evaluate the different integrated strategies
Description
Trabajo de Fin de Máster en Internet de las Cosas, Facultad de Informática UCM, Departamento de Ingeniería del Software e Inteligencia Artificial, Curso 2020/2021