Sistema de Identificación de Objetos en Espacios Cerrados Basado en Segmentación Semántica

Abstract

This project proposes to perform object recognition in indoor environments based on semantic segmentation using deep neural networks. To achieve this, two reference architectures widely used in the vision field have been selected: YOLO (YOLOv7) and Mask R-CNN. The choice of the YOLOv7 architecture for object detection is due to its ability to efficiently identify objects in real time. YOLO uses a single pass through the neural network to detect objects in an image, making it especially suitable for real-time applications where speed is a critical factor. On the other hand, the Mask R-CNN architecture was selected to address the task of semantic segmentation. It is used for mask assigment to each detected object, which provides detailed information on the precise shape of each object in the image. Both architectures, YOLO and Mask R-CNN, have been trained and evaluated using the well- known COCO (Common Objects in Context) database. COCO provides a wide variety of labeled and annotated images, allowing neural networks to be trained on a diverse set of object cate- gories and contexts.