Simulación de escaneados 3D
Archivos
NO SE HA AUTORIZADO la consulta de los documentos asociados
Fecha
2020-11-12
Autores
Título de la revista
ISSN de la revista
Título del volumen
Editor
Jaén: Universidad de Jaén
Resumen
Las nubes de puntos 3D procedentes de un sensor LiDAR se utilizan en un número considerable de aplicaciones, sin
embargo, más allá del coste de su adquisición, se observa un reducido número de conjuntos de datos etiquetados, que a
menudo contienen pocas clases, e incluso se asignan manualmente.
La simulación de un sensor LiDAR sobre un escenario 3D modelado permite obtener nubes de puntos sintéticas,
correctamente etiquetadas, con clases ajustadas a un escenario concreto, y con un nivel de detalle personalizado. Además,
la generación de un gran número de nubes de puntos puede obtenerse mediante la introducción de escenarios
procedurales.
Por otro lado, el comportamiento físico del sensor representa una elevada carga de trabajo. Por tanto, la introducción de la
computación paralela puede ayudar a reducir el tiempo de respuesta del escaneo. Por último, la simulación del sensor
también debe introducir aquellos errores más comunes vinculados a un dispositivo LiDAR.
3D point clouds given by LiDAR sensors have many applications nowadays. Beyond their acquisition cost, there is only a small number of labelled point clouds. Furthermore, labels are commonly reduced to a few classes, which also affect their level of detail. They may even be assigned manually, which suggest there could be errors in the labelling process. Therefore, simulating a LiDAR sensor on a 3D modelled scenario allows creating synthetic points clouds, properly annotated with classes which are defined for a specific scene and a customized level of detail. Moreover, generating a large number of points clouds can be achieved with procedural scenarios. An accurate simulation of a LiDAR sensor is a time-consuming task. Thus, parallel computation is included in this work to reduce the response time of the scanning process. Finally, simulating a LiDAR sensor also involves simulating errors which are related to such a sensor.
3D point clouds given by LiDAR sensors have many applications nowadays. Beyond their acquisition cost, there is only a small number of labelled point clouds. Furthermore, labels are commonly reduced to a few classes, which also affect their level of detail. They may even be assigned manually, which suggest there could be errors in the labelling process. Therefore, simulating a LiDAR sensor on a 3D modelled scenario allows creating synthetic points clouds, properly annotated with classes which are defined for a specific scene and a customized level of detail. Moreover, generating a large number of points clouds can be achieved with procedural scenarios. An accurate simulation of a LiDAR sensor is a time-consuming task. Thus, parallel computation is included in this work to reduce the response time of the scanning process. Finally, simulating a LiDAR sensor also involves simulating errors which are related to such a sensor.