Dataset of images for visual and non-visual analysis of colour applications in architecture

Abstract

This paper describes three datasets which include 443 folders and approximately 4430 images. The images were obtained from the interior of a 1:50 scale model using a fisheye camera connected to a Raspberry Pi microcomputer. This dataset aims to analyze the photobiological effects (visual and non-visual) of the interplay between coloured surfaces and different types of lighting strategies. The experiments were conducted under three types of light sources: simulated daylight through a mirror-box artificial sky simulator, direct daylight, and an electric lighting system that allows for colour temperature modification. This dataset includes low dynamic range images to generate high dynamic range images, which in turn can be used to plot false colour maps concerning photopic luminance, melanopic luminance, CCT of an image, M/P ratio, and brightness distribution maps. This dataset can be useful for architects, interior designers, and building engineers to integrate lighting and colour strategies according to the visual and non-visual needs of the users. This research was partially used in the research of Espinoza-Sanhueza et al. [1,2]. The datasets are published and shared through a Mendeley repository [3].

Specifications Table

Subject	Architecture
Specific subject area	The datasets contribute to calculating and analyzing photobiological effects of surface colour applications, daylight, electrical light, colour temperature of light, and lighting design.
Type of data	Images (Raw, analyzed)
Data collection	Three imagery datasets were obtained from a 1:50 physical model that represented a generic room. The ambiances differ in terms of surface colour application, type of light source, and lighting design position in relation to the point of view of the observer. The images were acquired using a fisheye lens (RPICamera-I by Waveshare) attached to a Raspberry Pi microcomputer [4] as observed by Parsaee et al. [5]. The lenses are characterized by a fixed aperture of f/2, a focal length of 1.55mm, and a 5-megapixel OV5647 sensor with a CCD size of 1/4inch which could produce images with 2592-by-1944-pixel resolution. The fisheye lens has a diagonal angle of view of around 185-degree as described in WaveShare [6]. Low-dynamic range images were obtained from very dark exposure values to very bright. The LDR images were set at ISO-100 and fixed white balance (D65). A Python script using OpenCV [7] libraries and ExifTool [8] is developed to generate HDR images from LDR images. The HDR images were calibrated for photobiological analyses based on Jung and Inanici [9], Parsaee et al. [5] and enhanced by Bolduc et al. [10].
Data source location	School of Architecture, Laval University, Quebec City, QC, Canada. Latitude and longitude: [46° 48′ N, 71° 12′ W]
Data accessibility	Repository name: Dataset of images for visual and non-visual analysis of colour applications in architecture DOI: 10.17632/hc7f9hnhxy.1 Direct URL to data: https://data.mendeley.com/datasets/hc7f9hnhxy/1
Related research article	The datasets were partially used in the following paper: •C. Espinoza-Sanhueza, M. Hébert, J.-F. Lalonde, C.MH. Demers, Biophilic analogous patterns for light-responsive architecture during polar night: Examining the photobiological effects of electrical lighting on surface colour configuration, Building and Environment 249 (2024) 111125 http://doi.org/10.1016/j.buildenv.2023.111125

1. Value of the Data

• •These datasets can be employed to evaluate the impact of colour on surfaces, daylight and electrical systems, and lighting design according to photobiological lighting conditions (vision and circadian stimulation) in indoors.
• •This data can be useful for architects and lighting designers to implement colour in architecture while assuring a proper light dose for biological responses.
• •This dataset could serve as a valuable reference for enhancing the decision-making process regarding the implementation of light and colour strategies in architecture.
• •The data can be used for future perception studies to analyze emotional responses according to colour and lighting design strategies in architecture.

2. Background

This dataset was created to study surface colour and architectural light as a potential solution to meet the photobiological needs of humans considering the photoperiod of subarctic and arctic latitudes. Photobiological effects of light were divided into visual considering photopic units to support vision, and non-visual considering melanopic units to enhance circadian stimulation [11]. Prolonged periods of darkness, also called polar night, prevent humans from receiving the light and spectral intensity necessary to perform visual tasks or to synchronize the circadian clock. High light intensity during summer, also known as polar day and midnight sun, can overstimulate the alertness state of individuals, generating sleep disorders. Considering that the colour properties of interior surfaces can modify the spectral properties of light, this paper presents a dataset of the photobiological effects of achromatic, monochromatic, and multicoloured combined spaces under different lighting design strategies. This dataset uses analogue techniques such as the use of reduced-scale models under direct sunlight (daylight) and advanced techniques such as LED electrical systems and a mirror-box sky simulator which mimics the daylighting conditions of a northern sky. The data acquisition was performed using a fisheye camera and HDR technique to obtain results comparable to what the human eye perceives.

3. Data Description

This article presents 443 architectural spaces displayed as images and divided into three main folders that compose the dataset. The datasets are shared through a Mendeley Data [3] repository based on surface colour applications and lighting strategies, including 1-Daylight-simulated, 2-Direct-daylight, and 3-Electrical. The dataset comprises LDR images that allow HDR image generation, false colour maps in photopic, melanopic, M/P ratios, CCT units (using the CIE algorithm), and brightness false colour maps. The scenes under simulated daylight are divided into a reference folder (0-Reference) and another four folders according to the colours applied in the surfaces: 1-Blue, 2-Green, 3-Red, and 4-Yellow. Each colour folder is also distributed according to the surface colour application (i.e., 1-Front wall, 2-Side wall, 3-Floor, 4-Ceiling, 5-Wall floor ceiling), and then classified according to its finish, Glossy or Matte, and colour temperature of the simulated daylight conditions: 2700 K, 4500 K, and 6500 K. The scenes under direct daylight and electrical lighting systems are also divided into the colour applied to the surfaces mentioned in Fig. 1. Acronyms were identified in Table 1 for colour combinations. Different from the multicoloured surface colour applications, the monochromatic scenes are split into subfolders according to 1 or 5 applied colour surfaces. Each surface colour application scene was then subdivided according to its lighting strategy position identified by SL (Side Lighting only for direct daylight analyses), GL (General Lighting for electrical systems), STL (Side Toplighting), FTL (Frontal Toplighting), and ZTL (Zenithal Toplighting). The scenes under electrical lighting were tested with white (5500 K), warm-white (4000 K), and warm (2600 K) colour temperatures identified in the data repository as subfolders of 5500, 4000, and 2600 respectively.

Fig. 1.

Data tree and folder descriptions.

Table 1.

Acronyms description of data tree and folders.

Acronyms	Description
BO	Blue-Orange The first colour refers to the floor, second colour refers to the frontal wall from the POV of the observer.
OB	Orange-Blue The first colour refers to the floor, second colour refers to the frontal wall from the POV of the observer.
VY	Violet-Yellow The first colour refers to the floor, second colour refers to the frontal wall from the POV of the observer.
YV	Yellow-Violet First colour refers to the floor, second colour refers to the frontal wall from the POV of the observer.
BRY	Blue-Red-Yellow The first colour refers to the floor, the second colour refers to the frontal wall, and the third colour refers to the side right wall.
VOG	Violet-Orange-Green The first colour refers to the floor, the second colour refers to the frontal wall, and the third colour refers to the side right wall.

Each folder contains from 1 to 11 LDR images from very dark exposure values (e.g., -3EV) to very bright exposure values (e.g., +3EV). A subfolder named Analysis_Results contains the HDR files in the company with the tone-mapped images and false colour maps for photopic, melanopic and brightness units. An example of the files content of each scene is exemplified in Fig. 2.

Fig. 2.

Example of the folder content of 0-Reference-6500 ambiance, including the LDR images, and the Analysis_Results subfolder that contains the HDR file, tone-mapped images, and false colour maps for photopic, melanopic, M/P Ratio, CCT units, and posterized brightness maps.

4. Experimental Design, Materials and Methods

The datasets were created using reduced-scale models under simulated daylight, direct daylight, and an electric lighting system. Two prototypes were used in the context of the presented data set. A first reduced-scale model was created for the 1-Direct-Daylight dataset. It presented dimensions similar to a generic room (10 m x 7 m x 3 m), constructed at 1:50 resulting in a box measuring 50 cm x 35 cm x 15 cm with a 90% WWR. The model allowed changes in terms of colour, percentage of colour in the space (i.e., surface colour configuration), and finish. The images were captured from a viewpoint set at the back of the room. The simulated daylight was generated using a mirror-box artificial sky simulator, with the dimensions of 2.4 m x 2.4 m x 2.4 m configured at ∼3800 lux horizontally as illustrated in Fig. 3. The device is equipped with 40 tuneable LED lamps permitting balancing warm and cool luminaires. The lamps are installed on the top of the cabin, with a 24.1 cm to acrylic diffuser, which accompanied by mirror acrylics on its interior surfaces generate infinite reflections that simulate overcast conditions. CCT properties of the artificial sky used in this dataset correspond to the daylight conditions found during the equinox periods of fall, early spring, and spring seasons of an Arctic sky exemplified in Fig. 3.

Fig. 3.

a) Mirror-box artificial sky simulator composition. b) CCT comparison of the artificial sky and real northern skies used in the sub-data set 1-Daylight-simulated. c) Experimental setup and illustration of distance control for image capture. Retrieved from Espinoza-Sanhueza et al. [2].

The models were installed in an experimental table in the centre of the device. The hues used in the datasets of simulated daylight were created using coloured cardboard in blue, green, red, and yellow. The coloured cardboard applied in the study area served to create four surface colour (SC) configurations: frontal (10% SC), sidewall (15% SC), floor (30% SC), ceiling (30% SC), and a colour configuration covering almost entirely the space except for the window on the left-hand side wall (85% SC). Finally, two types of finishes were compared: matte and glossy. The matte surface finish was simulated using coloured cardboard with its regular texture. For the glossy surface finish, a transparent plexiglass sheet was applied to offer a highly reflective coating. The colour information in the CIELab colour system, glossiness properties and variable combinations are mentioned in the Fig. 4.

Fig. 4.

a) Variable combination for sub-data set folder 1-Daylight-simulated. Retrieved from Espinoza-Sanhueza et al. [2].

The images under direct-daylight and electrical lighting were obtained from a second reduced-scale model consisting of a plywood prototype manufactured in the carpentry laboratories of the Laval University School of Architecture, serving as a support shell for the experimental study area. The support shell corresponded to a top-opened box of 50.6 cm × 36.2 cm x 20 cm fabricated with 6 mm plywood. It is assembled with a steel corner exo-structure and fastened with 5/8′′ wood screws. The panels, that configure the experimental space, were inserted into the shell and created the study area representing a generic room of 50 cm × 35 cm × 15 cm. Fourteen different colour concepts were studied as illustrated in Fig. 1: achromatic spaces including white, and black; monochromatic spaces including blue, red, yellow, orange, green, and violet; and multicoloured spaces including the combinations of blue and orange, violet and yellow, and triad combinations of blue-red-yellow and violet-orange and green. The monochromatic spaces were also studied according to 1 surface and 5 surfaces applied in the space (Fig. 5).

Fig. 5.

a) Lighting position and surface colour configuration variables for sub dataset 2-Direct-daylight and 3-Electrical. B) Spectral properties of coloured panels and light sources.

Four lighting scenes according to their luminaire position were implemented from the observer's point of view: General lighting (GL), Side Toplighting (STL), Frontal Toplighting (FTL), and Zenithal Toplighting (ZTL). The tests under direct daylight were performed under clear sunlight conditions during May 2022. The experimental setup was installed in the backyard of the School of Architecture of Laval University (46.81, -71.20) as presented in Fig. 6. The reduced-scale models presented a south-east orientation. The datasets developed under electric lighting used a lightbox of the dimensions of 50.6 cm × 35 cm x 10 cm that fixed the LED stripes. The electrical system was installed at 10 cm from the top of the lightbox to the ceiling panel and placed on the top of the experimental area to generate a diffuse light in the experimental room. The light intensity of the simulated electric system was set at 1000 lux and three different CCTs: warm at 2600 K, warm white at 4000 K and white at 5500 K. These settings were verified using a CL-200A Konica Minolta Chromameter [12]. This equipment is composed with a light sensor that capture illuminance levels (lux) and correlated colour temperature (CCT) measured in kelvin degrees (K). The advantage of measuring these parameters with the CL-200A is that a relationship can be created between the effects of coloured surfaces and light with the colour perception and illumination levels computed from HDR images.

Fig. 6.

Experimental setup for sub-dataset 2-Direct-daylight and 3-Electrical. 1) refers to the exostructure used to support the prototype shell. 2) Support shell. 3) Interior panels. 4) Removable ceiling according to the lighting strategies. 5) Light box that supported the electrical system. Retrieved from Espinoza-Sanhueza et al. [1].

4.1. Capture of LDR images

To prompt the image capture, a low-cost 185° field of view fisheye Camera Module attached to a Raspberry Pi 4 was introduced in the physical model as exemplified in Fig. 6. The Raspberry Pi comprised a Python script to initiate the image capture. The R-Pi was remotely controlled using a Virtual Network Computer (VNC) server to avoid movement during the image and data acquisition. The Pi Camera was furnished with a Python script which permitted to obtaining LDR images from very dark (e.g., -3EV) to very bright exposure values (e.g., +3EV) as displayed in Fig. 7. The different exposure values are modified by the shutter speed of the camera. The aperture size of the camera remains fixed at F-1.8. The Python script was configured to set a white balance of daylight and ISO 100 for all the images. LDR images were stored in sRGB colour space (i.e., red, green, and blue) and .JPG format.

Fig. 7.

LDR image collection and HDR generation.

4.2. HDR images generation

HDR image generation was performed using a Python script based on the research of Parsaee [5] and enhanced by Bolduc et al. [10] using OpenCV [7] and Exiftool [8] libraries. HDR images were generated by covering camera response functions (CRFs) based on Devebec and Malik [13] as mentioned in the dataset provided by Parsaee et al. [14], offered as an OpenCV method.

4.3. Light intensity calibration

Light intensity for photopic, melanopic, M/P Ratio, and CCT units were calibrated through a script developed by Bolduc [15]. This script offers a geometric calibration of the camera to project an orthographic projection as exemplified in Fig. 8. This procedure permits computing illuminance levels between scenes from the centre of the images and neglects the corners of the image characterized by black pixels. Further specifications and equations are presented in the paper of Bolduc et al. [10] Section 9.

Fig. 8.

Geometric model.

4.4. Tone-mapped imaged and photobiological false colour maps

HDR images were used to plot tone-mapped images and photobiological false colour maps in photopic, melanopic, M/P Ratio, and CCT units as displayed in Fig. 9. HDR files contain luminance values which means that typical screens are not capable of displaying the luminance levels of a real screen. To address this shortcoming, tone mapping operators were created to compress the original luminance values into a manageable dynamic range that screens can reproduce and safeguard crucial image details. In this experiment, we have used Reinhard [16] tone-mapped operators since studies from Yoshida et al. [17] indicated that individuals appreciated higher naturalness and higher details in bright regions. Photobiological false colour maps were plotted using a script from Parsaee et al. [14,18] by calibrating RGB and XYZ (CIE tristimulus) channels of HDR images based on the chromatic calibrations explained in Jung and Inanici [9]. The employed equations and procedures are illustrated in the research from Parsaee et al. [14] page 10.

Fig. 9.

Photobiological false colour maps of photopic, melanopic, M/P Ratio and CCT units.

4.5. Brightness false colour maps plot

Tone-mapped images obtained from HDR images served to plot brightness distribution maps employing the greyscale analysis developed by Demers [19] to evidence lighting patterns or attention points of brightness. Tone-mapped images were imported in Adobe Photoshop [20] and treated using a greyscale filter to standardize the brightness levels of each pixel. A posterization filter using a 4-degree scale for simulated daylight and a 5-degree scale for direct daylight and electrical sub-dataset was applied as specified in Fig. 10.

Fig. 10.

Brightness contrast data treatment.

The described approach is suitable for extended lighting analyses involving HDR images and post-processing. It enables the calculation of photobiological effects through images and Python coding as a common language between architects and the scientific community in computing vision.

Limitations

Not applicable.

Ethics Statement

The authors confirm that the current work does not involve human subjects, animal experiments, or any data collected from social media platforms.

CRediT authorship contribution statement

Carolina Espinoza-Sanhueza: Conceptualization, Methodology, Software, Data curation, Formal analysis, Validation, Visualization, Investigation, Writing – original draft. Marc Hébert: Supervision, Funding acquisition, Writing – review & editing. Jean-François Lalonde: Supervision, Funding acquisition, Writing – review & editing. Claude MH Demers: Supervision, Funding acquisition, Writing – review & editing.

Data Availability

• Dataset of images for visual and non-visual analysis of colour applications in architecture (Original data) (Mendeley Data).