Research Projects

Current projects:

Funded by the German Research Foundation (DFG) – 445336968

The current research project deals with developing an AI-based approach to the determination of dynamically changing, user-oriented luminaire control curves based on continuously recorded sensor input values to realize HCL-compliant system solutions for LED interior lighting. The term “control curve” can be used to summarize all those parameters that can be varied for each luminaire largely independently of one another and with time. For the planned investigations, these are the horizontal illuminance Eh generated on the work surface, the proportion of indirect light in this illuminance (according to Houser), the most similar color temperature (CCT), and the indirect blue light proportion, which significantly defines the physiologically activating component of the lighting and is to be described in the present research project with the help of the circadian stimulus (CS) value (according to Rea and Figueiro).

To determine optimal, AI-based control curves, test subject tests are carried out at different times of the day to determine the users' respective individual light preferences depending on both subjective-psychological and external influencing factors and to correlate them with the corresponding sensor data with the help of machine learning methods. The aim is to develop a lighting system that, based on the incoming sensor data and using an underlying user preference model, can predict the optimal control curve for the current situation and dynamically adjust the room's lighting accordingly over the course of the day.

As a starting point for the implementation of the planned research project, support points of individual control curves are initially recorded by recording user preferences related to the current light situation in the room at different times over several working weeks as part of a first test person study and using AI-supported evaluation with the simultaneously recorded sensor data Relation to be set. Based on this data, the support points determined in this way are summarized in individual, characteristic clusters depending on the ambient conditions as well as the subject-specific influencing factors (e.g., age, gender, chronotype, current state of mind, etc.), which correspond to similar lighting preferences of the test persons. In a second iteration step, these clustered support points are then presented to the participating test subjects in a further test subject study and further adapted to the individual preference using various setting options made available to the test subjects. Together with the continuously recorded sensor data, these additional input values further optimize the control curves. In the third part, a physiological evaluation of these optimal control curves in terms of maximizing user preference is to be carried out in the context of a further study by test persons to answer the question of their activating effect and finally to derive a corresponding user preference model as the basis for future, intelligent lighting systems.

Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – 450942921

The research project deals with the development of an AI-based control of the automotive light distribution to generate situationally optimized, dynamic light distributions. The aim of this research project is to break away from the conventional division of light distributions into low beam and high beam and to present holistic light distributions. In order to develop these light distributions, extensive studies will be carried out to record the current traffic area. Here, representative data for the German traffic area will be recorded first. The aim of the planned drives is to record the German traffic area in its entirety, taking into account different road classes analogous to their real occurrence. For the evaluation, the latest algorithms for the recognition of objects are trained individually and geometric distributions of different objects, cars, trucks, buses (each driving and parking), traffic signs (depending on class), pedestrians, cyclists are created. From these object distribution data as well as other recorded data such as speed, road class, road conditions etc. different traffic situations are automatically created. In addition to the recording of the German traffic area, contrast investigations and luminance analyses in the different road classes are carried out. These serve to determine the safety-relevant light requirements for motor vehicle drivers and to record the current light conditions in the German traffic area. Not only the recognizability distance and the necessary contrast are considered, but also the perception of brightness and homogeneity of the foreground illumination. This is essential, since in addition to an objective increase in safety, the subjective perception of safety also has a strong influence on the driver's well-being and behavior. The results of these investigations are also incorporated into the optimization of the situation-dependent light distributions. In addition, a further study records the gaze behavior of the drivers and calculates optimized light distributions by combining the object distributions and the gaze behavior in road traffic. These theoretical light distributions are then validated. In a first step, the various light distributions are displayed in a driving simulator, independent of the technology used, in the various virtually generated traffic situations. The acceptance and the feeling of safety of the test persons is tested virtually when using the different light distributions. In addition, a further study will test the extent to which drivers' visibility changes as a result of the newly generated light distributions.

Funded by the German Research Foundation (DFG) – 450636577

The eye's pupil diameter is an essential factor in photometric and visual examinations due to its direct influence on retinal illuminance and retinal image quality. A smaller pupil diameter can ensure a more considerable depth of field and a reduction in optical aberrations of the lens, which can positively affect the eye's visual acuity. Various studies show that the optimal pupil diameter for visual tasks in the photopic luminance range is between two and three millimeters. With today's technology of multi-channel LED lights, it is possible to optimize light spectra to directly influence the pupil opening, color and brightness perception, or other visual metrics. The number of different LED color channels in such a system determines the wheel of freedom, allowing visual metrics to be kept constant while others can be changed. The first step towards actively optimizing the pupil opening through interior lighting without influencing other imaging visual parameters is developing a valid pupil model that predicts the spectral and time-dependent pupil diameter of humans. To characterize and model the human pupil behavior, a temperature-controlled 15-channel LED lamp was developed at the lighting technology department (see Figure 1).

Existing pupil models are based on the L- and M-cones' sensitivity in the retina, making them unsuitable for achieving an ideal pupil opening in a light spectrum optimization process [1]. Using these models would make it impossible to vary the pupil diameter while maintaining the brightness and color perception parameters. There is currently a significant prediction inaccuracy when using these pupil models due to the time independence and luminance use as a V (λ) -weighted photometric parameter. The integration of a V (λ) -weighted photometric variable in a pupil model leads to significant prediction errors of the pupil diameter when using multi-channel LED lights as a light source [1] (see Figure 2). This prediction error's main reason is the wavelength-dependent sensitivity curve of the pupil, which does not correspond to the V (λ) curve alone [1,2]. This effect was only understood by discovering the intrinsically photosensitive ganglion cells (ipRGCs), which contributed to the fundamental understanding of stimulus processing in the retina, the perception of brightness, the circadian rhythm, and the pupillary light reaction. These neurophysiological findings have so far been insufficiently integrated into the existing pupil models. The consequence of this are prediction errors of the absolute pupil diameter of current pupil models (see Figure 2).

In our funded DFG research project, we aim to develop a new pupil model based on deep learning, which considers the time dependency and the adaptive weighting of the retinal receptors to control pupil behavior. The latest findings on the mechanism of pupil behavior are to be covered with this model. Specifically, we would like to combine existing time-variant and time-invariant model approaches with a data-driven, non-parametric model to develop an overall model of pupillary behavior. With this, we take into account the dynamic and spectrally dependent receptor adaptation of the phasic and tonic pupillary behavior. The non-parametric model approach ensures a constant improvement in the model quality as the database increases. This is to be achieved through a publicly accessible pupil database that can also be edited by external working groups. The basic model and the basic data should be developed in this research project employing extensive empirical pupil examinations and serve as a general impetus for this generic model project.

The concept for developing a pupil model, which integrates the current findings from basic neurophysiological research, would enable the time-dependent calculation of the pupil diameter for different light spectra for the first time. The accurate calculation of the spectrally and time-dependent pupil diameter is essential in visual experiments, eye safety calculations, and medical diagnostics. Such a model can also help to show the relationship between the perception of brightness and the pupil control path [3].

References

[1] Zandi B, Klabes J, Khanh TQ. Prediction accuracy of L- and M-cone based human pupil light models. Sci Rep 2020;10:10988. https://doi.org/10.1038/s41598-020-67593-3.
[2] Zandi B, Kunst K, Khanh TQ. Einfluss der melanopsinhaltigen Ganglienzellen auf die kurz- und langzeitige Pupillenlichtreaktion. 120. Jahrestagung der Dtsch. Gesellschaft für Angew. Opt., Darmstadt: 2019.
[3] Zandi B, Guo X, Bodrogi P, Khanh TQ. EXPERIMENTAL EVALUATION OF DIFFERENT BRIGHTNESS PERCEPTION MODELS BASED ON HUMAN PUPIL LIGHT RESPONSES. Proc. CIE 2018 Top. Conf. SMART Light., vol. 2, International Commission on Illumination, CIE; 2018, p. 201–8. https://doi.org/10.25039/x45.2018.OP34.

Data collection and storage form the basis for tapping previously unused innovation potential for lighting systems with new use cases. However, the luminaires or lighting systems, including the associated sensors, must record data, transfer them to a cloud with uniform semantics and save them. In this project, the data from test systems installed across 14 locations are brought together in a common IoL cloud. This project aims to analyze the sensor and luminaire data generated in the test systems while simultaneously testing the architecture for data storage and data transmission. Building on this, the centrally stored data is analyzed and evaluated, and data-driven lighting use cases are examined and processed.

The leading research questions for the “Internet of Light” project as a joint project between the German lighting industry and the research institutions are:

  • How can the data generated by sensors and lights in the lighting systems be transmitted, persisted, and analyzed in a uniform, clearly structured data format efficiently and with low latency in one or more cloud systems?
  • How do you deal with transmission failures, physically implausible data, and restrictions concerning the system?
  • Which sensor types and operating data are required for data-driven use cases in different lighting systems?
  • What mutual interactions, influences, and correlations do the luminaire and sensor data have with one another?
  • What benefits can users and system operators gain from the luminaire and sensor data over longer periods of several seasons, for example, to optimize and prove economic efficiency, environmental friendliness, and user-friendliness under real laboratory conditions and for later field systems?
  • Which algorithms and methods can be used for processing sensor data in the field of lighting technology?

 

 

 

Visual Efficiency – A sustainable evaluation scale for modern architectural and lighting design

Current architectural lighting design covers only the minimal values given by legal norms securing the possibility of visual tasks. This leads to a lack of acceptance, reduced well-being and sustainability. Our approach introduces a new evaluation scale called “Visual Efficiency” that expands the classical pure visual evaluation by including psychophysical and psychological factors, user preferences as well as energy efficiency. To develop such a new evaluation scale, field studies intended to provide access to both quantitative and semantic rating data of architectural lighting situations will be conducted.

Eventually, this will pave the way for achieving the vision of a general architectural lighting guide that includes all requirements to generate well-being for the user and offers a versatile, sustainable and efficient usage.

The project is part of a collaboration with the FG Entwerfen und Gebäudetechnologie of FB 15 (Architecture).