This paper presents a novel computational imaging technique that leverages alternating current (AC) illumination from electrical grids. By passively detecting AC lighting oscillation patterns, the approach enables identification of "the type of bulbs in the scene, the phases of the electric grid up to city scale, and the light transport matrix". This reveals previously undetectable scene properties and enables applications including reflection unmixing, semi-reflection analysis, nocturnal high dynamic range imaging, and scene re-rendering under different lighting conditions. The authors develop a coded-exposure HDR imaging method optimized for AC grid lighting and compile the DELIGHT database — "the first of its kind in computer vision" — containing temporal lighting response functions for various bulb types.

Artificial lighting powered by alternating current (AC) is ubiquitous in urban environments. The AC power supply causes light sources to oscillate at characteristic frequencies — typically 100 Hz or 120 Hz (double the grid frequency of 50 Hz or 60 Hz). Different bulb types (incandescent, fluorescent, LED) respond to AC with distinct temporal signatures: incandescent bulbs oscillate smoothly due to thermal inertia, fluorescent tubes exhibit sharper modulation, and LEDs may show complex patterns depending on their driver circuits. This paper's key insight is that these oscillations, normally considered a nuisance (flicker), encode rich information about the scene's lighting structure.

By capturing a short video sequence and analyzing the per-pixel temporal frequency content, one can extract: (1) the light transport matrix — which light source illuminates which scene point; (2) bulb identification — what type of light source is present; (3) grid phase mapping — which electrical phase powers each light, enabling city-scale infrastructure analysis. The approach requires only a camera and a power outlet (for synchronization), operating in completely uncontrolled outdoor environments across different countries and voltage standards (110V and 220V).

Prior work on light transport analysis typically requires active, controlled illumination — projecting structured light patterns to decompose direct and indirect illumination. Temporal multiplexing methods use high-speed projectors and cameras in laboratory settings. Photometric stereo requires known, calibrated light positions. In contrast, this work achieves light transport analysis passively, in uncontrolled environments, using existing AC lighting as the "projector". The AC grid effectively provides temporal coding for free — a natural structured light system that covers entire cities.

The method models the time-varying appearance of a scene under AC illumination. Each pixel's intensity is a superposition of contributions from multiple light sources, each with its own temporal response function (TRF). The TRF encodes how a specific bulb type converts AC electrical power into light output over time. By capturing video at sufficient frame rate and performing frequency analysis (Fourier decomposition), the per-pixel contributions from different light sources can be separated. The DELIGHT database catalogs measured TRFs for dozens of bulb types, serving as a spectral dictionary for source identification.

A key technical contribution is the coded-exposure HDR imaging method specifically designed for AC lighting. Standard HDR imaging uses multiple exposures at different durations, which is suboptimal for oscillating light sources. The proposed method uses exposure timing synchronized to the AC cycle, capturing images at specific phases of the oscillation. This enables nocturnal HDR imaging with a single standard 30 Hz camera — the ACam prototype. By choosing exposure windows that sample different parts of the AC cycle, the method effectively brackets the dynamic range within a single AC period.

The framework enables several compelling applications. Light transport decomposition: by identifying which light source contributes to each pixel, the scene can be re-rendered with specific lights turned on/off, achieving virtual relighting without any controlled capture. Reflection separation: when a glass surface creates semi-reflections, the two layers (transmitted and reflected) are often illuminated by different light sources with different AC phases, enabling passive separation based on temporal signatures. Grid phase mapping: in three-phase power systems, different lights operate on different phases (120 degrees apart), and this phase information can be extracted from video to map electrical infrastructure across city blocks.

Experiments demonstrate the method's capabilities across multiple countries and voltage standards (110V in North America, 220V in Europe and Asia). The DELIGHT database successfully characterizes dozens of bulb types with distinct temporal response functions. Nocturnal HDR imaging at distances spanning meters to kilometers achieves high dynamic range capture using the ACam prototype. Light transport decomposition successfully separates contributions of individual light sources and enables convincing virtual relighting results. Bulb classification achieves reliable identification even in mixed-lighting scenarios. The system's operation requires only a camera synchronized to a power outlet and a largely stationary scene.

This paper demonstrates that the ubiquitous AC power grid provides a naturally occurring structured illumination that encodes rich scene information. By passively analyzing temporal oscillations in AC-powered lighting, a standard camera can extract light transport matrices, bulb types, and electrical grid phases — information previously accessible only through active, controlled illumination setups. The DELIGHT database and coded-exposure HDR method provide the foundations for future work in this direction. The approach opens new possibilities for computational imaging in everyday, uncontrolled environments at scales ranging from individual rooms to entire cities.