“This projects aims to propose a system to extract trajectories performed by people in a closed space through videos from a camera network. This system is composed by four components: (i) people identification, (ii) people re-identification, (iii) generation of trajectories, and (iv) visualization of trajectories where components (i) and (ii) develop the video processing. On the one hand, identification was carried out using a pre-trained model called Mask RCNN based on regional convolutional neural networks. On the other, re-identification employes XQDA, which is a method to learn a subspace of the image characteristics with the aim og separating images from different people and gathering images which belong to the same person. The system was tested with videos of the General Library of the Andes University getting results consistent with the methods used”
Este evento pretende mostrar los proyectos y desarrollos que se realizan en las Universidades en temas relacionados de visión por computador y procesamiento de lenguaje natural.
#OrgulloDISC #YoSoyDISC #ComunidadDISC #TodosSomosDISC