Article Dans Une Revue Journal of Real-Time Image Processing Année : 2023

Optical flow algorithms optimized for speed, energy and accuracy on embedded GPUs

Résumé

Embedded Computer Vision is a hot field of research that requires trade-offs in order to balance execution time, power consumption and accuracy. In that field, dense optical flow estimation is a major tool used in many applications. Many algorithms have been designed, focusing on accuracy, very few works address trade-offs and implementation on embedded hardware. This paper tackles these trade-offs for embedded GPU through the example of the well-known TV-L 1 algorithm. Thanks to High Level Transforms-operator fusion and pipeline-and taking into account the iterative aspect of these algorithms, we achieve a speedup of ×3.7 versus OpenCV. Moreover, we show that a 16bit half precision implementation has a higher accuracy than the 32-bit precision one for the same frame processing time on NVIDIA Jetson boards. Furthermore, this work can be generalized to any kind of iterative stencil-based algorithms.

Fichier principal
Vignette du fichier
JRTIP21_OpticalFlow-1.pdf (1.46 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-04199828 , version 1 (08-09-2023)
hal-04199828 , version 2 (11-09-2023)

Licence

Identifiants

Citer

Thomas Romera, Andrea Petreto, Florian Lemaitre, Manuel Bouyer, Quentin L. Meunier, et al.. Optical flow algorithms optimized for speed, energy and accuracy on embedded GPUs. Journal of Real-Time Image Processing, 2023, 20 (2), pp.32. ⟨10.1007/s11554-023-01288-6⟩. ⟨hal-04199828v1⟩
1177 Consultations
1113 Téléchargements

Altmetric

Partager

  • More