Inverse light estimation: Estimating the light of outdoor scenes through textures
Translated title
Omvendt lysestimering: Estimering af lyset af udendørs scener igennem teksturer
Author
Therkelsen, Mikael Nørlund
Term
4. term
Education
Publication year
2024
Submitted on
2024-05-28
Pages
39
Abstract
This project explores estimating the lighting of an outdoor scene to support augmented reality (AR). The system uses textures (surface images) obtained from the Unity game engine to infer how the scene is lit. The estimated lighting is then used to shade a virtual object inserted into the scene so it blends in more naturally. Method: A lighting model was implemented in Python. The Python program takes Unity textures as input to estimate the scene’s light settings. It connects to the Unity project via sockets and sends the estimated values across the connection. Results: Performance was mixed. The estimated values did not match the actual lighting values, but they tracked changes in lighting as expected. In some lighting conditions, the estimates were sufficient to help the inserted object blend into the scene. Conclusion: The system shows promise but needs further development to improve accuracy and stability.
Dette projekt undersøger, hvordan man kan estimere lysforholdene i en udendørsscene for at understøtte augmented reality (AR). Systemet bruger teksturer (billeder af overflader) fra Unity-spilmotoren til at udlede, hvordan scenen er belyst. De estimerede lysindstillinger bruges derefter til at skygge et indsat virtuelt objekt, så det falder mere naturligt ind i scenen. Metode: En belysningsmodel blev implementeret i Python. Python-programmet tager Unity-teksturer som input og anslår lysindstillingerne. Programmet er forbundet til Unity-projektet via sockets og sender de estimerede værdier gennem forbindelsen. Resultater: Systemets ydeevne var blandet. De absolutte estimater matchede ikke de faktiske lysværdier, men de fulgte ændringer i belysningen som forventet. I nogle lysforhold var estimaterne gode nok til, at det indsatte objekt blandede sig ind i scenen. Konklusion: Systemet viser potentiale, men kræver mere arbejde for at forbedre nøjagtighed og stabilitet.
[This apstract has been rewritten with the help of AI based on the project's original abstract]