Today, aerial camera systems for photogrammetry are diverging into two main categories: systems for photogrammetry, and specialized systems for content programmes. The traditional photogrammetry approach is modernized but still with the same basic demands for cameras and processing as in the 1980s. For content, the diversity is significant with camera systems and processing both advancing strongly. The number of specialized camera systems already exceeds that of traditional cameras, and they often feature simpler workflows focused on fast, frequently fully automated processing.

Inspired by the human eye

At a fundamental level, all camera systems are inspired by the structure of the human eye. They consist of three core components: the image sensor (analogous to the retina), the lens (with the aperture mimicking the iris) and intelligent processing, where software interprets and extracts meaning from captured imagery (comparable to the brain’s visual cortex).

With this framework in mind, this article examines each of these elements, considers future developments and provides some practical guidance. The focus is specifically on airborne camera systems. While satellites and drones are also important imaging platforms, they involve significantly different operational conditions and will be addressed separately.

The image sensor

From the very beginning, digital cameras have been able to match the performance of analogue systems. The first images from the Vexcel UltraCam D were truly impressive, marking a significant leap forward in quality. It also introduced simultaneous near-infrared (NIR) capture, which was a major advantage. At that time, the pixel size was 9µm. Today, pixel sizes have been reduced to less than one third of that, while achieving significantly higher sensitivity.

Photogrammetric cameras are complex systems that merge data from multiple sensors – colour, monochrome and NIR – handled through sophisticated and demanding post-processing into a single eye view. In contrast, modern camera modules have largely standardized sensor technology, most notably the Sony IMX811, which offers 19,200 × 12,800 pixels. A single compact unit can now do what a photogrammetric camera system could do ten years ago. So the question is, do we need to follow the old way with complicated stitching or can we do something better?

In any camera, image quality is influenced by three main settings: ISO, shutter speed, and aperture. The SONY IMX811 sensor can handle the ISO setting electronically. The global shutter sensor IMX661 can, in addition to ISO, handle shutter speed electronically. Combined with a fixed aperture, this results in a fully electronic camera. This is an important next step, as such cameras can capture not just one image, but several with different settings in milliseconds. Gpixel delivers a true alternative to the abovementioned image sensors, but all camera manufacturers currently use the same IMX series.

The lens

Along with its mechanical shutter and aperture, the lens is often the most overlooked components of aerial cameras, yet also one of the most technically challenging. As image sensors improve, the demand on the lens increases linearly with the growing number of pixels.

While the optics are the key element in utilizing the increased number of pixels, this seems to be the hardest element for customers to understand. Optics have improved significantly, with better glass, coating and manufacturing, but merely assessing the classic values for lens quality – such as the maximum aperture – is no longer enough. To truly evaluate the quality,  the lens must be considered in combination with the image sensor and software.

As a physical object, the lens is highly influenced by temperature and pressure. The longer the focal length, the more it is affected, which is why classical calibration remains essential. Lens distortion, even in the best lenses, requires corrections of 50-150 pixels to achieve metric accuracy, and both eccentricity and focal length vary with temperature. The most stable parameter is the distortion pattern (see Figure 1), and the final result remains stable within half a pixel.

Every single camera unit needs calibration, and full-system calibration is needed if several cameras are used. Single-head calibration is typically done in a laboratory, while system calibration is done in the air when the system is ‘warm’ (operational for at least 20 minutes).

In addition to the optics, lenses also include mechanical components such as the aperture and shutter. These are largely based on designs originating from the first analogue cameras. The most common shutters, such as Prontor, have a maximum shutter speed of 1/500s and have been produced in almost the same way for decades. If these two mechanical components could be eliminated, lenses could be smaller, more stable and capable of transmitting more light.

New combinations of image sensors and lenses

The 250MP Sony sensors and similar models are a significant improvement from the 150MP sensors but are still very traditional in that their rolling shutters expose the sensor line by line. In contrast, global shutter sensors (e.g. IMX661) offer significant new possibilities as they expose the full image at once. Besides enabling electronic control of ISO, they also feature electronic shutters with speeds up to 1/16,000s.

Another important advantage of global shutter sensors is their ability to capture multiple images of the same object in milliseconds. This makes it possible to use different ISO and shutter speed settings for each image – one optimized for sharpness and another for maximum colour information, for instance – and then combine them. Blur between the images is no longer a problem, even when captured by the fastest aircraft, and image quality is significantly improved in low-light conditions.

The aperture is the only remaining issue to be addressed to achieve a fully electronic camera system. A practical approach to this would be to fix the aperture at a single value and control the amount of light using the shutter speed. This capability can also reduce the need for a dedicated camera operator, as automatic camera settings become easier to implement.

Image processing

Image processing and adjustment cannot be avoided, but information is lost each time. It is better to say as close to the raw information as possible. Therefore, processing would ideally be limited to a final adjustment of coverage.

The traditional large-format camera consists of up to 12 individual sensors capturing images that are combined into a single image after several advanced adjustments, including calibration. In other words, what is often seen as a single image is far from raw data. Moreover, this is a time-consuming post-processing step that is not possible until after landing. If the single images were to be used from IMX811 or IMX661, for example, they could be processed into a final image on the fly, so the results would be ready upon landing. The new generation of AI processing units could even perform advanced adjustments, enabling the images to be used immediately for purposes such as emergency response.

What’s next?

How aerial images are utilized is changing fast, but traditional photogrammetry is developing slowly. Mesh modelling is increasingly taking over and Gaussian Splatting seems to be the next step. The cameras need to support all these technologies and AI-driven feature extraction. In existing image sensors, we can expect the trend towards more pixels to continue. However, as the major markets for sensors, the focus has shifted to commercial and industrial use and partly for AI.

For aerial mapping, the most important development will be access to fully electronic cameras. They are more electronically stable, more affordable and achieve better performance. Moreover, electronic cameras are more compact. They are also the key to a simplified workflow that retains more basic information from images, making it easier to use machine learning and AI. We are there almost with automatic production of large coverages with a high accuracy; now, we need AI-ready images to enable detection of precisely the right information to meet a specific need.

The human eye is truly impressive, and cameras designed as an ‘electronic eye’ have made a lot of sense so far. But it is time for a different way of thinking. For instance, a fly’s eye (Figure 3) consists of thousands of different sensors, enabling it not only to see beyond our visual spectrum, but also to sense temperature, judge speed and navigate in darkness.

The first step towards this approach could be a freer and easier merging of current image sensors. Sensors with different resolutions, angles, lenses, etc., could be combined more freely using AI’s new processing capabilities. The next step would be to combine sensors outside the visual spectrum (e.g. Lidar, thermal), covering much more than what the human eye can see. This will take time, but it is important to set the wheels in motion now. If the industry only further improves traditional aerial cameras, the future will be challenging. A good starting point would be flexible systems that leverage modern sensors for both photogrammetry and broader content capture.