Share

A luminance-based approach for inferring personal daylight preferences using a new composite similarity index.

Number: 3488

Author(s) : XIONG J., TZEMPELIKOS A., MAH D.

Summary

The concept of extracting similarity features between two luminance maps is presented in this study for classifying lighting preferences of office workers. A composite similarity index is proposed to quantify the similarity between different lighting conditions, using processed luminance maps and luminance contrast maps, captured from High Dynamic Range (HDR) images in real offices together with occupant visual preference data. 7 existing lighting parameters were selected for evaluating the performance of the similarity index. Logistic regression models were trained using each parameter separately to classify the occupant’s visual preference. A zoning approach of the generated similarity maps was developed for a fair comparison and to prevent information loss during averaging. Similarity indices in each zone of both maps were then averaged and used as input parameters of a logistic regression model. The results showed that: (i) positive and negative signs of similarity indices were necessary for describing the occupant’s visual preference; (ii) dividing the similarity maps into 3 zones showed better classification performance; (iii) both luminance similarity and luminance contrast similarity maps are important for classifying the occupant’s visual preference; and (iv) similarity index maps could show more stable classification results than existing parameters. This composite similarity index scheme can be used in more complex deep learning techniques to predict and learn occupant’s visual preferences.

Available documents

Format PDF

Pages: 10 p.

Available

Free

Download

Details

  • Original title: A luminance-based approach for inferring personal daylight preferences using a new composite similarity index.
  • Record ID : 30030243
  • Source: 2022 Purdue Conferences. 7th International High Performance Buildings Conference at Purdue.
  • Publication date: 2022
  • Document available for consultation in the library of the IIR headquarters only.

Links


See other articles from the proceedings (39)
See the conference proceedings

Indexing