2026-05-05T02:01:50.226Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:1:10080 2024-07-01T14:12:30Z 1

Segmentation of digital images using deep neural networks for accurate vehicle damage recognition 2023 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/1/10080 П. Бишевац П. Спалевић Р. Ивковић Vehicle accidents are a public occurrence that often results in vehicle damage. The capability to rapidly and accurately detect and categorize the extent of the vehicle damage can save time and reduce costs for insurance firms and companies who deal with vehicle repairing. In this manuscript, we propose a different approach to vehicle damage recognition using deep neural networks for image segmentation. We primary trained our model on a dataset of vehicle images with known damage. The resulting neural network was then applied to a novel set of digital images, and the segmented parts of damage were compared to ground truth annotations. Our tests display that our method achieves high accuracy in detecting vehicle damage parts. The proposed method has the possible to advance the productivity of vehicle damage assessment and repair processes, and can be simply joined into current vehicle inspection workflows. Additional developments can be made by incorporating supplementary data sources such as contextual information and other meta data. conferenceObject LXVII KONFERENCIJA ETRAN