skip to main content
Idioma:

A study of the effect of noise and occlusion on the accuracy of convolutional neural networks applied to 3D object recognition

Garcia-Garcia, Alberto ; Garcia-Rodriguez, Jose ; Orts-Escolano, Sergio ; Oprea, Sergiu ; Gomez-Donoso, Francisco ; Cazorla, Miguel

Computer vision and image understanding, 2017-11, Vol.164, p.124-134 [Periódico revisado por pares]

Elsevier Inc

Texto completo disponível

Citações Citado por
  • Título:
    A study of the effect of noise and occlusion on the accuracy of convolutional neural networks applied to 3D object recognition
  • Autor: Garcia-Garcia, Alberto ; Garcia-Rodriguez, Jose ; Orts-Escolano, Sergio ; Oprea, Sergiu ; Gomez-Donoso, Francisco ; Cazorla, Miguel
  • Assuntos: 3D object recognition ; Caffe ; Convolutional neural networks ; Deep learning ; Noise ; Occlusion
  • É parte de: Computer vision and image understanding, 2017-11, Vol.164, p.124-134
  • Descrição: •The accuracy of CNNs for 3D object recognition with noise and occlusion is studied.•Volumetric representations (space subdivision and occupancy measures) are studied.•Both 2D (with 3D input) and 3D CNNs with pure 3D convolutions were used.•The ModelNet database was converted to point clouds with noise and occlusions.•Results confirm that volumetric representations play a key role on the robustness. In this work, we carry out a study of the effect of adverse conditions, which characterize real-world scenes, on the accuracy of a Convolutional Neural Network applied to 3D object class recognition. Firstly, we discuss possible ways of representing 3D data to feed the network. In addition, we propose a set of representations to be tested. Those representations consist of a grid-like structure (fixed and adaptive) and a measure for the occupancy of each cell of the grid (binary and normalized point density). After that, we propose and implement a Convolutional Neural Network for 3D object recognition using Caffe. At last, we carry out an in-depth study of the performance of the network over a 3D CAD model dataset, the Princeton ModelNet project, synthetically simulating occlusions and noise models featured by common RGB-D sensors. The results show that the volumetric representations for 3D data play a key role on the recognition process and Convolutional Neural Network can be considerably robust to noise and occlusions if a proper representation is chosen.
  • Editor: Elsevier Inc
  • Idioma: Inglês
Voltar para lista de resultados

  • Realização: ABCD-USP USP   Logos de Redes Sociais: Freepik

Buscando em bases de dados remotas. Favor aguardar.