hide
Free keywords:
-
Abstract:
1. Motivation, specific objective: The world around us is complex in its visual attributes, differing in spectrum, intensity, colour, spatial make-up and temporal properties. Natural scenes have long been known to be statistically regular, and several publicly available data sets use calibrated or uncalibrated sensors to analyse these statistical regularities. However, no dataset currently captures the spectral, spatial and temporal properties of natural scenes comprehensively. To characterise the world around us comprehensively vis-à-vis its visual and non-visual physiologically- relevant properties, we started an image-capture campaign to map out the spectral, spatial and temporal properties of both indoor and outdoor natural scenes across a wide range of geographical and seasonal contexts. 2. Methods: We developed a multi-modal, minimal-parallax image capture setup for capturing natural scenes comprehensively. We are collecting radiance images using an α-opic imaging radiometer (40°x48° FOV, 2712x3388 px 2 ), spectral irradiance measurements using a high- resolution spectroradiometer (380-780 nm, 1 nm resolution), illuminance and colorimetric (xy) measurements (~8 Hz), depth information using a depth camera (15 fps), and uncalibrated wide-field RGB videos (60 fps). All measurement devices, controlled by a laptop are integrated in a compact box (42x32x27 cm 3 ), which can be mounted on a tripod. Power is supplied through external batteries, making the system suitable for both indoor and outdoor use. In addition to the primary light-based data, we comprehensively describe the scenes using a novel metadata schema containing location, weather and other information. We collect natural scene data using two distinct protocols. First, in a time-lapse protocol, we collect data from the same scene over the course of a day. Second, in a trajectory protocol, we are simulating a hypothetical trajectory of an individual throughout the day, thereby capturing plausible natural scenes that people might be exposed to throughout the day. 3. Results: We have finished piloting and technical checks and are now collecting time-lapse data across various natural scenes. We have subjected our image-capture setup to a rigorous cross- validation check, indicating high between-device agreement for related quantities. Our setup has been proven to be portable for use in field measurements, facilitating the development of a geographically unbiased, worldwide database, which we plan to develop over time and make available. 4. Conclusions: Our data collection campaign will yield highly significant real-world data for developing novel approaches in understanding illumination and scenes in the real world. All data captured with our system are converted to physiologically relevant, cone-, rod- and melanopsin-based quantities and metrics to relate our data to lighting recommendations and guidelines, thereby creating the first open-access reference resource for the spectral, spatial and temporal properties of natural scenes.
