Results

System and Services

VEDI: Vision Exploitation for Data InterpretationFarinella G. M., Signorello G., Battiato S., Furnari A., Ragusa F.,

Leonardi R., Ragusa E., Scuderi E., Lopes A., Santo L., Samarotto M.

DMI – IPLab, CUTGANA, University of Catania

Xenia Gestione Documentale s.r.l. – Xenia Progetti s.r.l.

http://iplab.dmi.unict.it/fpv

References

Method F1 score

57-POI 0,59

57-POI-N 0,62

9-Classifiers 0,66

Proposed 0,68

[1] Ragusa F., Furnari A., Battiato S., Signorello G., Farinella G. M.,

Egocentric Visitors Localization in Cultural Sites, in ACM Journal on

Computing and Cultural Heritage (JOCCH), 2019.

[2] Ragusa F., Furnari A., Battiato S. , Signorello G., Farinella G. M.,

Egocentric Point of Interest Recognition in Cultural Sites. In 14th

International Conference on Computer Vision Theory and

Applications (VISAPP), Prague, Czech Republic, February 25-27,

2019.

[3] F. L. M. Milotta, A. Furnari, S. Battiato, M. De Salvo, G.

Signorello, G. M. Farinella. Visitors Localization in Natural Sites

Exploiting EgoVIsion and GPS. In 14th International Conference on

Computer Vision Theory and Applications (VISAPP), Prague, Czech

Republic, February 25-27, 2019.

[4] Website: http://iplab.dmi.unict.it/VEDI_project/

VEDI is an integrated system which includes a wearable device capable of

supporting the visitors of cultural sites, as well as a back-end to analyze the

visual information collected by the wearable system and infer behavioral

information useful for the site manager.

Architecture

The VEDI system is made up of 4 components: 1) Mobile

devices, 2) GPU, 3) Charging and update station, 4)

Central system.

HoloLens GoPro

mFF1 0,82 0,81

mASF1 0,71 0,71

Datasets

Localization and PoIs Recognition on UNICT-VEDI

Monastero dei Benedettini Orto Botanico

- 9 Environments (temporal annotations)

- 57 Points of Interest (54248 frames with temporal and

bounding boxes annotations)

UNICT-VEDI

UNICT-VEDI_Succulente

EgoNature

- 16 Points of Interest (36728 labeled

images)

- 9 Environments (temporal

annotations and GPS locations)

Accuracy Time (ms)

SqueezeNet-6 + DCT 0,86 4,7

SqueezeNet-9 + DCT 0,86 6,09

SqueezeNet-11 + DCT 0,86 6,60

SqueezeNet + DCT 0,91 22,9

Localization on EgoNature

Method F1 score

AlexNet 0,89

PoIs Recognition on UNICT-VEDI_Succulente

http://iplab.dmi.unict.it/fpvhttp://iplab.dmi.unict.it/VEDI_POIs/