Best practices for image acquisition and photogrammetry

This article describes the best practices for capturing a dataset to obtain high-qualityresults.

Photogrammetry

Photogrammetry is the science of taking accurate measurements of physical objects from photos to precisely define its shape, dimensions, and position in 3D space. To work with photogrammetry is needed: 

Images: Photogrammetry is based on the principle of stereoscopy, which is similar to human vision, where each eye captures an image of the same object from two different viewpoints slightly separated by distance. For photogrammetry, this means the images are captured with an overlap, frontal (between subsequent images) and lateral (between adjacent images or flight lines) from different positions. Whlie the primary input is overlapping images, it is important to know the internal and external parameters of the cameras to effectively process the data captured:

• Internal: Internal parameters define the geometry of the camera including the focal length, principal points X and Y, and distortion parameters (R1, R2, R3, T1, T2). 
• External: The external parameters refer to the position and orientation of the camera for each captured image.

Each image is composed of pixels. The Ground Sampling Distance (GSD) is the distance between the centers of two adjacent pixels (from center to center) measured on the ground. The GSD determines the amount of detail in each image and is dependent on the megapixels of the camera and the distance between the camera and the object.

To improve the accuracy of photogrammetric reconstructions, it is strongly recommended to use Ground Control Points (GCPs). A GCP is a point with known coordinates in the area of interest, which can be measured using traditional surveying methods such as total stations or GNSS receivers. While GCPs are not mandatory for project processing, they significantly enhance the absolute accuracy of the results. Additionally, GCPs can be utilized as Checkpoints to verify the accuracy of the results. Checkpoints are the most precise way to confirm the accuracy of the reconstructions.

Acquisition plan

The first step planning in a photogrammetry project is the acquisition plan. A good dataset is required to generate high-quality and accurate results. To obtain a good dataset, follow the steps below:

1. Design the Image Acquisition Plan considering the following:

2. Configuring the Camera Settings

Whenever possible, the camera settings should be set to automatic and remain consistent throughout the image capture. At times, the camera settings used to acquire the images need to be configured. Wrong configuration can result in images with blur, noise, distortions, etc.

• Camera body and lens
  • Perspective and fisheye lenses are supported.
  • The zoom not be changed.
  • A consistent fixed focal length is recommended.
• Camera settings
  • Internal stabilization settings should be off
  • Shutter/aperture/ISO should be on automatic
  • If the images are blurry or noisy, manually set shutter/aperture/ISO
  • Set the Manual Focus to Infinity

3. Georeferencing the Images

The images can be georeferenced using a camera with a built-in GPS or using RTK/PPK workflows.

4. GCPs (optional, but recommended)

Using GCPs requires planning how many GCPs are required, as well as where and how they will be measured.

Where should the GCPs be placed? 

• To ensure consistent accuracy, the GCPs should be uniformly distributed across the area of interest. They should also be as close as possible to objects of interest and offset at least one flight line from the edge of the project area.

How accurate should the GCPs be, and how should they be measured? 

• The accuracy of the GCPs depends on the accuracy needed for the final results. It should correspond to the final absolute accuracy required for the project. The GCPs can be measured using a variety of topographic equipment:
  • Total stations can reach millimetric-level accuracy.
  • GNSS receivers can reach centimeter-level accuracy (depending on the equipment, project location, and country).

How can a GCP be identified on the flight?

• The GCP target must be visible in the images. A GCP target (as seen in the image below) should be five to ten times the dimensions of the GSD. For example, if you have a 2-centimeter GSD, the target should be a minimum of 10 to 20 centimeters to be clearly identified in the images, though often a slightly larger target is beneficial. 

5. Weather conditions

Weather conditions, especially light, precipitation, and atmospheric clarity can either improve or degrade the quality of photogrammetry results. To minimize weather-related issues, it's often recommended to plan photogrammetry surveys during favorable weather and lighting conditions.

• Light. The best lighting for photogrammetry is soft and consistent, which can be achieved with artificial light or cloudy skies. This reduces shadows and ensures clear, even illumination, making it easier to capture details. Avoid direct sunlight, direct sunlight creates harsh shadows and overexposure, which can hide details and cause reflections.
• Wind. Avoid strong wind. While wind is not only a concern for safety reasons, it can also affect the sharpness of images. Strong winds can cause motion blur, especially when using drones. For safety and image clarity, calm weather conditions are preferred.
• Rain. Rain can lead to water droplets on lenses, diminishing image clarity. It can also obscure certain features and change the surface conditions, making it more challenging to capture accurate data.

6. GSD 

Ground Sampling Distance GSD is a critical factor that influences the accuracy of the data processing and the quality of the outputs. Essentially, it defines the resolution of the imagery and influences the level of detail that can be captured.

Horizontal accuracy is usually 1 to 2 times the GSD and vertical accuracy is 2 to 3 times the GSD, so it is very important to calculate the GSD according to the project needs.

Use cases

The ideal image acquisition plan depends on the type of terrain/object to be reconstructed and the desired results: 

• General case: For projects that do not include forests, snow, lakes, agricultural fields, and/or other terrains that are difficult to reconstruct.
• Forest and dense vegetation: For a project with areas covered by forest or dense vegetation.
• Flat terrain with agriculture fields: For flat terrain with homogeneous visual content such as agriculture fields.
• Building reconstruction: For 3D modeling of buildings.
• Special cases: For snow, sand, and water surfaces (oceans, lakes, rivers, etc).
• Corridor mapping: For projects with a linear area of interest (roads, rivers, etc).
• Multiple flights: For projects with images taken using multiple flights.
• City reconstruction (visible facades): For 3D modeling of urban areas.
• 3D interior reconstruction: For 3D modeling of the interior of buildings
• Mixed reconstruction: For combined datasets (interior/exterior and/or aerial/terrestrial and/or nadir/oblique).
• Large vertical objects reconstruction: For 3D modeling objects that are tall and slender.

General case 

The recommended minimum overlap for most cases is at least 75% frontal overlap (with respect to the flight direction) and at least 60% side overlap (between flight lines). It is recommended to take the images with a regular grid pattern. The camera maintains a constant height over the terrain/object to ensure the desired GSD.

Forest and dense vegetation

Forest and dense vegetation often look very similar in overlapping images due to its complex structure, which includes thousands of branches and leaves. As this is a very complex environment, it is recommended to modify the General case with the following:

• Increase the overlap between images to at least 85% frontal and side overlap.
• Increase the flight height: Flying at a higher altitude reduces perspective distortion and increase the visual characteristics of vegetation, which can make it easier to detect visual similarities in overlapping images.
• Avoid flying conditions where the vegetation is moving significantly in subsequent images.

Flat terrain with agricultural fields

In cases where the terrain is flat with homogeneous visual content such as agriculture fields, it is difficult to extract common characteristic points (keypoints) between the images. To achieve good results, it is recommended to use a single grid mission similar to the one in General cases with the following changes:

• Increase the overlap between images to at least 80% frontal and side overlap.
• Fly higher. Flying higher can improve the results as long as the required GSD is maintained.
• Have accurate image geolocation.

Building reconstruction

Reconstructing 3D buildings requires a specific image acquisition plan, known as an "orbit" or "circular" mission:

• Camera pitch. There is no universal camera angle suitable for all projects. Position the tilt to ensure most of the image frame captures the object to be reconstructed.
• Fly a second and third time around the building increasing the flight height and decreasing the camera angle with each round.
• It is recommended to take one image every 5-10 degrees to ensure sufficient overlap, depending on the size of the object and its distance to it.

Special cases

This section presents some hints for areas that are difficult to map such as terrains with snow, sand, lakes, etc.

• Snow and sand have little visual content due to their homogeneous nature. Therefore:
  • Increase the high overlap: At least 85% frontal overlap and at least 70% side overlap.
  • Set the exposure settings to get as much contrast as possible in each image.
• Water surfaces have almost no visual content due to being homogeneous, dynamic, and reflective. Sun reflection on the water and waves present difficulties for visually matching the same object in different images.
  • Open water is nearly impossible to reconstruct.
  • To reconstruct other water surfaces, such as rivers or lakes, each image needs to have identifiable features, such as land. Flying higher may help to include land or other identifiable features.
  • If the project area includes water, it is essential that at least three-quarters of each image contains identifiable features other than water.

Corridor mapping

Mapping a corridor such as a railway, road, or river requires at least two flight lines, though at least three are recommended. GCPs are not required but are recommended to improve the accuracy of the reconstruction.

When flying a corridor with two flight lines, it is recommended to use at least 85% frontal overlap and 60% side overlap. Nadir images or oblique images can be used. For flat terrain, nadir images are recommended

Multiple flights

It is possible to process a single project using images captured from multiple flights. When designing the different image acquisition plans, make sure that:

• Each plan captures the images with sufficient overlap.
• The images are captured under more or less the same conditions (sun direction, weather conditions, no new buildings, etc.).

City reconstruction (visible facades)

The 3D reconstruction of urban areas requires a double grid image acquisition plan so that all the facades of the buildings (north, west, south, east) are visible on the images. 

• Overlap should be 85% frontal overlap and at least 70% side overlap.
• Camera tilt. Images should be taken with an angle between 10º to 35º to capture the facades. For most double-grid missions, an angle of 70º is recommended. 
• If more detail is needed, combine the aerial imagery with terrestrial images. In such cases, it is strongly recommended that GCPs or Manual Tie Points be used to merge images captured with different image capture methods.

3D Interior reconstruction

Terrestrial images are strongly recommended for interior reconstruction. A high overlap (90%) is needed.

Mixed reconstruction

It is possible to combine interior/exterior, aerial/terrestrial, and/or nadir/oblique.

Ensure there is sufficient overlap within each dataset and between datasets. In such cases, it's highly recommended to use GCPs or Manual Tie Points to properly combine the different datasets.

Large Vertical Object reconstruction

The 3D reconstruction of objects that are tall and slender requires a specific image acquisition plan:

• Fly close to the structure, maintaining a safe operating distance,
• Turn several times around the structure at different heights.
• Images should be taken with a high overlap: 90% overlap between images taken at the same height and 60% overlap between images taken at different heights.
• Everything in the image frame must be in focus, including objects in the background that are outside the project area.
• Having image geolocation is recommended.