Detection and classification of road and objects in panoramic images on board the ATLASCAR2 using Deep Learning

• Autor(a) principal: Costa, Rúben Daniel Ferreira da
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/31452
• topic: Autonomous driving; Deep learning; Panoramic images; Image segmentation; Road and object detection

The field of autonomous driving has been increasingly explored and the future of transport partly depends on the use of this type of vehicle. For an autonomous car to navigate on the public road, it must be able to detect everything around it, ensuring that the actions taken do not compromise the safety of any person. Within the Atlas project, this dissertation aims to create a model that allows the detection of road and objects in panoramic images, thus increasing the ATLASCAR2 field of view. In view of this need, a system was developed for the creation of panoramic images through images acquired by the cameras mounted on the car and, to make the detection of the road and other objects, deep learning was used to train the models in order to ensure great accuracy and detail in detection. This work presents the results obtained with the trained models, presenting a comparison between the use of different architectures and datasets. In addition, an evaluation of the capacity of these models was also performed in the city of Aveiro.