Diplomovka

vedúci: RNDr. Miroslav Opiela, PhD.

konzultant: RNDr. Šimon Horvát, PhD.

https://s.ics.upjs.sk/~martina_kuchtova/

Ciele

Investigating current neural network-based methodologies for indoor positioning.
Introducing a neural network model for indoor positioning utilizing primarily camera imagery and outlining the procedure for video data acquisition.
Evaluating the accuracy of the proposed method within a selected facility, examining its performance under image modifications (e.g., blurring) and in the context of the positioning task (classification versus regression).
Exploring the potential integration of the proposed model with other neural networks or its incorporation into an existing indoor positioning system.

 References
(Recommended reading)

DEADLINES

Timeline

NázovDátumyStatusPoznámky
Untitled


Data anonymisation

Methods

To domethod "name"previewstatusaffectednotesmethod typeEnviroment/dependenciesResult
face swap (Real Time)scriptedfaceRealTime ntb cameramodelVS Codedlib
Example
face swap scriptedfaceworks on img, videomodelVS Codedlib
Example
LAMAmodelDocker
haven' tested yet
pixel (face)scriptedfaceOpenCVcenzorhaar_cascadesENV23
Example
censored (face)scriptedfaceblack rect - profile&front face / eyes basedOpenCVcenzorhaar_cascadesENV23
Example
blur (person)in processbodyOpenCVblurhaar_cascadesENV23
haven' tested yet
blur (face)scriptedfaceprofile&front face / eyes basedOpenCVblurhaar_cascadesENV23
Example
Untitled
haven' tested yet
Untitled
haven' tested yet
censor (face) domcoldominant colour rectangle
haven' tested yet
remove heads/faces
haven' tested yet
remove people
haven' tested yet
replace face (overlay)
haven' tested yet
blur (whole image)
haven' tested yet


Anonymisation candidates:

  • face
  • head (face & hair)
  • upper body
  • body
  • image region
  • whole image

Potential solutions:

➳ hair, moustage, glasses
➳ hood
➳ face filters, face-swap
➳ removal (AI supported)

Data

NN v. 0.1

empty hall trained; 3 segments (2 of them are following parts of the hallway)

Dataset

NameCMAccuracynotesDateStatus
Empty hall (train)99.72
Empty hall empty_confmatrix.png96.33
People presentpeople_confmatrix.png49.33