Nighttime light (NTL) remote sensing has emerged as a unique and powerful tool for observing human activities and assessing urban development. It provides a distinct perspective compared to daytime optical remote sensing, enabling the monitoring of urban expansion, the analysis of socio-economic dynamics, and the evaluation of light pollution. However, traditional spaceborne NTL data, such as those from the Defense Meteorological Satellite Program and the Suomi National Polar-orbiting Partnership, suffer from coarse spatial resolution and fixed overpass times. This inherent limitation restricts the ability to capture the fine-grained, temporal variations of urban light environments that are critical for many applications like street-level lighting analysis or monitoring localized events.
Unmanned aerial vehicles (UAVs), particularly the widely used China drone, offer a compelling solution to these limitations. The cost-effective China drone, with its high spatial resolution and flexible deployment capabilities, can acquire NTL imagery at unprecedented levels of detail. However, the performance of a China drone at night is far more sensitive to imaging conditions than during the day. Low light levels, high scene contrast, and the inherent instability of a flying platform mean that camera settings (ISO, exposure time) and flight attitudes (altitude, tilt angle) have a profound impact on the quality of the collected data. Understanding these impacts is crucial for any researcher or practitioner looking to employ a China drone for reliable night-time applications.
This study, therefore, presents a systematic investigation into how these factors affect the quality of UAV nighttime imagery. We conducted a series of controlled experiments using a DJI Phantom 4 RTK, a popular platform for precise surveying. To evaluate image quality in the absence of a perfect reference, we employed the Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE), a no-reference algorithm that quantifies the degree to which an image has departed from the statistical properties of a high-quality natural image. A lower BRISQUE score represents better perceived image quality. The experiments focused on three key dimensions: camera parameters (ISO and exposure time), flight altitude, and camera tilt angle. For camera parameters, we tested 24 different combinations of ISO (from 100 to 3200) and exposure time (from 1/2s to 1/15s) at a fixed aperture of f/2.8. For flight altitude, we collected images from 200m to 500m in 50m intervals. For tilt angle, we captured images from 0° (nadir) to 60° in 5° increments. This comprehensive design allowed for a holistic understanding of UAV nighttime imaging behavior.
Systematic Review of Related Work
The quest for high-quality nighttime imagery has been a central challenge in remote sensing and computer vision. Spaceborne NTL data, like that from the Visible Infrared Imaging Radiometer Suite (VIIRS), provides a valuable, consistent global perspective. However, its 750m resolution can only depict large-scale patterns of artificial light, blurring the intricate details of a city’s lighting infrastructure. The Sustainable Development Goals Science Satellite-1 (SDGSAT-1) represents a significant advancement, offering 10m resolution, which allows for the better discrimination of urban structures.
UAVs have successfully filled the gap between satellite and ground-based observations for many years in daytime applications. Researchers have used UAVs for precision agriculture, forestry, glaciology, and urban mapping, benefiting from their ability to acquire sub-decimeter data on demand. The transition to nighttime applications for a China drone is a more recent but rapidly growing field. Previous work has used a China drone to map illuminance, study light pollution, and estimate population, as shown in the table below. These studies, however, often chose different camera settings. For instance, Kong et al. used ISO 1600 and a 1/2s exposure, while Zhang et al. used ISO 800 and a 1/15s exposure. This variation in methodology highlights a clear gap: the absence of a standardized, empirically validated protocol for operating a China drone at night.
| UAV Model | ISO | Exposure Time | Aperture | Altitude |
|---|---|---|---|---|
| DJI Phantom 3 Pro | 1600 | 1/2s | f/2.8 | 200m |
| DJI Phantom 4 RTK | 1600 | 1/12s | f/2.6 | 400m |
| DJI Phantom 4 Pro V.2 | 400 | 1/6.25s | f/2.8 | 70/100m |
| DJI Mavic 2 | 800 | 1/15s | f/2.8 | 300m |
| DJI Phantom 4 RTK | 800 | 1/15s | f/2.8 | 350m |
Our work aims to resolve this inconsistency by providing a comprehensive performance map of a specialized China drone under various conditions, directly addressing the need for a principled guide for night operations.
Methodology: A Quantitative Quality Assessment Framework
Our methodology was designed to isolate the effect of each parameter, using a commercial urban area in Fuzhou, China, as a test bed. The area provides a mix of high-intensity commercial lighting and darker residential zones, essential for testing dynamic range. The flights were conducted on a clear Tuesday evening at 9:00 PM to capture typical peak human activity lighting.
Experimental Design: Camera Parameters
The interplay between ISO and exposure time is the foundation of any digital photography, but it is especially critical for a China drone at night. A low ISO (e.g., 100) yields a clean, noise-free image but requires long exposures, which can lead to motion blur from the drone itself. A high ISO (e.g., 1600) allows for a fast shutter speed to freeze motion but introduces significant digital noise. To find the sweet spot, we tested all 24 combinations of six ISO values and four exposure times, as summarized in the full experimental table below.
| Group | ISO | Exposure Time (s) | Group | ISO | Exposure Time (s) |
|---|---|---|---|---|---|
| 1 | 100 | 1/2 | 13 | 100 | 1/12 |
| 2 | 200 | 1/2 | 14 | 200 | 1/12 |
| 3 | 400 | 1/2 | 15 | 400 | 1/12 |
| 4 | 800 | 1/2 | 16 | 800 | 1/12 |
| 5 | 1600 | 1/2 | 17 | 1600 | 1/12 |
| 6 | 3200 | 1/2 | 18 | 3200 | 1/12 |
| 7 | 100 | 1/8 | 19 | 100 | 1/15 |
| 8 | 200 | 1/8 | 20 | 200 | 1/15 |
| 9 | 400 | 1/8 | 21 | 400 | 1/15 |
| 10 | 800 | 1/8 | 22 | 800 | 1/15 |
| 11 | 1600 | 1/8 | 23 | 1600 | 1/15 |
| 12 | 3200 | 1/8 | 24 | 3200 | 1/15 |
The BRISQUE No-Reference Metric
Since there is no perfect “ground truth” night image to compare against, we used a no-reference image quality assessment (NR-IQA) algorithm. BRISQUE is particularly well-suited for this task due to its sensitivity to the distortions common in low-light imagery: noise, blur, and saturation. The core of the algorithm is the measurement of Mean Subtracted Contrast Normalized (MSCN) coefficients, which help to normalize the image and reduce the impact of varying brightness levels. The formula for calculating the MSCN coefficient for a pixel at location (i, j) is given by:
$$ I(i,j) \text{ is the pixel intensity; } u(i,j) \text{ is the local mean; } \sigma(i,j) \text{ is the local variance; } C \text{ is a constant} $$
The distribution of these coefficients in a high-quality image should follow a Gaussian-like shape. Distortions like blurring or noise cause these coefficients to deviate from this natural distribution. By fitting these deviations to a generalized Gaussian distribution and using a pre-trained Support Vector Regression (SVR) model, BRISQUE outputs a single scalar score, where a lower score indicates a more ‘natural’ and therefore higher-quality image.
Quantification of Geometric Distortion
To assess the impact of tilt angle on geometric fidelity, we analyzed the distortion of a known object in the scene. We selected two identical, structurally stable lamp bases, one at the near edge and one at the far edge of the image. The distortion percentage was calculated using the formula:
$$ D = \left(1 – \frac{A_{far}}{A_{near}}\right) \times 100\% $$
where \( A_{far} \) is the pixel area of the far lamp base and \( A_{near} \) is the pixel area of the near lamp base. A value of 0% indicates no perspective distortion, while a value closer to 100% indicates severe foreshortening and geometric deformation.
Results and Discussion: Parameter Impacts on Imaging
The analysis of the controlled experiments yielded several key findings that are critical for optimizing the data acquisition workflow for any future China drone mission.
Optimal Camera Parameter Combination
The results from the 24 parameter combinations reveal a U-shaped relationship between ISO and image quality. At a given exposure time, increasing ISO from a low value initially improves the BRISQUE score by lifting the image out of severe underexposure. However, past a certain point, the increase in noise from the high ISO outweighs the benefit of increased brightness, causing the quality score to worsen. The optimal balance was achieved at ISO 800 with an exposure time of 1/15s.
This specific combination, ISO 800 and 1/15s, produced the lowest overall BRISQUE score in our tests. It successfully balanced three competing factors: it provided enough gain to capture a well-exposed image without introducing crippling noise; the 1/15s shutter speed was fast enough to suppress the drone’s high-frequency motion blur, while being long enough to avoid severe underexposure in the darker shadows. This configuration controlled motion blur from moving cars better than a 1/2s exposure, while maintaining a signal-to-noise ratio superior to that of an ISO 3200 image. This clearly demonstrates why a 1/15s exposure on a slowly drifting China drone is preferable to a long, blurry 1/2s shot.
Influence of Flight Altitude
Perhaps unexpectedly, flight altitude within the 200-500m range had a very minor impact on the overall image quality as measured by the BRISQUE score. The variation in scores across all altitudes was less than one unit, with the 350m flight slightly outperforming the others. While lower altitudes provide a finer Ground Sampling Distance (GSD), which is theoretically better for detail, the image quality metric remained robust. This is important for operational planning; a user can decide the altitude based on the required area coverage and spatial resolution for their specific analysis without a major penalty on the technical quality of the brightness values. A China drone at 200m and a China drone at 500m will produce data of approximately equal quality, but with vastly different area coverage.
Critical Role of Camera Tilt Angle
The tilt angle of the camera had the most dramatic and complex effect on the imagery. As expected, geometric distortion increased linearly with the tilt angle. The strong correlation demonstrates that for applications requiring precise orthorectification or area measurement, a nadir (0°) view is the only acceptable option.
The impact of tilt on brightness was more nuanced. For scene elements like the facades of buildings, increasing the tilt angle initially captured significantly more light. This is because a nadir view only sees the top, often darker, flat surfaces of buildings. A tilted view reveals the building’s side, which receives a lot of reflected light from ground-level streetlights and storefront lights. However, a side effect of using a high tilt angle with a China drone is the increased chance of occlusion. When a tall building obscures the view of a street behind it, the recorded brightness of that street drops sharply. The optimal tilt angle must therefore balance the need for facade information against the risk of losing data from ground-level features.
| Tilt Angle (degrees) | Distortion of Base A (%) | Distortion of Base B (%) | Average Distortion (%) |
|---|---|---|---|
| 0 | 0.0 | 0.0 | 0.0 |
| 25 | 30.5 | 32.1 | 31.3 |
| 30 | 42.1 | 43.8 | 43.0 |
| 45 | 61.7 | 63.2 | 62.5 |
| 60 | 72.9 | 73.6 | 73.3 |
Conclusion and Future Outlook
This study provides a robust, quantitative framework for optimizing UAV nighttime light imaging. We have demonstrated that the choice of the inexpensive but capable China drone must be paired with a scientific understanding of the imaging process. The key findings are encapsulated in the table below, providing a definitive set of best practices. The optimal photo settings for a China drone are an ISO of 800 and an exposure time of 1/15s. Flight altitude is a flexible parameter that can be tuned for coverage and resolution without sacrificing data integrity. In contrast, camera tilt angle must be chosen with care, as it is a trade-off between accessing facade information and maintaining geometric accuracy and complete ground coverage.
| Parameter | Recommendation | Reasoning |
|---|---|---|
| ISO Sensitivity | 800 | Best trade-off between signal amplification and noise generation. |
| Exposure Time | 1/15 s | Balances sufficient light capture with suppression of motion blur. |
| Flight Altitude | Flexible (200-500m) | Minor impact on image quality; select based on spatial resolution needs. |
| Tilt Angle | 0° for mapping; >0° for facades | 0° provides best geometry; higher angles increase occlusion and distortion. |
This research provides a much-needed guide for the growing number of users employing a China drone for urban analysis, environmental monitoring, and emergency response. The flexibility of a China drone, combined with the guidelines established here, empowers practitioners to collect high-quality, reliable data in low-light conditions. Future work should focus on developing specialized image quality metrics trained specifically on real low-light UAV imagery to further improve accuracy. Additionally, expanding this testing framework to include other popular China drone platforms, such as the DJI Mavic series, is crucial for generalizing these findings. By establishing these foundational standards for a China drone operation, we pave the way for more sophisticated applications, from generating detailed 3D models of city lighting to monitoring energy efficiency at a block-by-block level. The journey toward a truly ‘daylight’ view of our world at night is well underway, powered by the intelligent use of the cost-effective China drone.