Photographing techniques

Photographing techniques #

In this chapter, we will take a look at how we can produce 3d polygon meshes from photographs. The quality of the model is dependent on many factors and understanding how each of the factors affect the the final model can help in producing better quality models.

The very first step in the photogrammetric pipeline is to detect the features and build a point cloud of the model. So, the detection of features in the model should be the first rule of thumb to distinguish between a bad and good input photograph. And hence all the tweaking of the factors listed below is to optimize in gathering many detectable feature on the subject while not detecting the features in the background.

Although the automatic settings on the digital cameras is good enough for taking photographs for everyday situations, they are not good enough for photogrammetric reconstructions. This section describes what settings to keep in mind while photographing the target. The three most important camera settings shutter speed, aperture and the ISO. In photography, this is known as the exposure triangle.

Camera shutter speed #

The shutter speed is the time the camera lens is exposed to the scene. The lower the shutter speed, the longer the lens is exposed to the scene and vice-versa. Allowing the lens to get exposed for a longer time, produces brighter images since more light can enter the sensor. However, if the target or the camera is moving and the shutter speed is low, then target or any other object in the field of view can expose more than one part in the sensor and hence the images may come out blurry.

It is hard to detected features in a blurry images, and since the quality of the model is dependent on the number of detected features, the final output may suffer. Therefore, avoid taking blurry images and try to keep the images as sharp as possible. Increasing the shutter speed can help.

With a higher shutter speed, the exposure time is less and the object in the field of view get less time to expose more area in the sensor and hence, the images are less blurry. However, the images are darker as a result. If the subject is in bright light, then higher shutter speed is recommended. However, it is not always possible to have bright lights everywhere.

Sometimes it is not easy to keep a steady hand while taking photographs, especially when the shutter speed is too low to compensate for ill lit scenes. Since we are taking photographs of temples and heritage sites, the photographer has no control over the position of the sun and area that is not directly lit like interiors and shadows. In such cases, a tripod may be used and the control the camera using a remote than physically pressing the shutter release button that may disturb the camera. In case of drone photography, it is advisable to make sure that the drone’s motion is stabilized and completely still when the photograph was taken.

Camera aperture #

The aperture is the size of the hole that exposes the sensor. It is donut-shaped ring that can expand and contract. The f-number or f-stop controls how big of a hole you want on your aperture. The f-number is generally represented as F/#. A higher aperture like F/4 or F/2 open the aperture more widely and hence more light can enter the sensor, while a lower aperture like F/32 or F/64 allows less light. However, the most important property that the f-number changes is the depth of field.

The depth of field, controls the focus and blurriness of the background. A higher depth of field brings out more details in the background and a lower depth of field blurriest the background and makes the subject of the photographs to stand out more. For our case of photographing cultural heritage sites, we are more interested in the subject rather than the objects in the background. And the photogrammetric techniques that we employ uses feature detection to track the various parts of the model.

A photograph with details in the background can be picked up as a feature and tracked. While they do not provide any useful information to the algorithm, they take up space and computing resources. So, use a low depth of field or higher f-number for taking the photographs.

Camera ISO #

The ISO controls the sensitivity of the sensor. It is always recommended to keep the ISO as low as possible. A lower ISO is always better for increasing the dynamic range, and color accuracy while decreasing the chromatic aberrations. A lower ISO also makes sensor less sensitive and hence reduces the noise and graininess. This is important since a less grainy image makes it less likely for the feature detection algorithm to pick up noise. However, a less sensitive sensor needs more exposure to detect less bright areas. So, it all comes down to the balance between the ISO, shutter speed and the aperture.

Subject zoom and overlap #

The photogrammetric algorithm does not care if the entire subject is not visible in every single shot. Therefore, close up pictures that can cover many detectable features are desirable. However, for the algorithm to know the entire model, it is also advisable to take some pictures with the entire subject in frame as well as making sure that for two images, there is at least a 70% overlap of the detectable features.

Subject material #

A shiny, specular surface that reflects the surrounding is difficult for the algorithm to work with since the reflections of the surrounding objects may be detected as a feature of the subject. Many lose parts like hairs and grids are difficult to capture and vary wildly from scene to scene. Even with high overlaps, the features are not easy to detect and only confuse the algorithm to try select and track non-existent features. For cultural heritage sites, it is much more practical to hand model such parts.

Subject masking #

Isolating a cultural heritage site from it’s background is not possible and if the subject is big enough, then the depth of field may not help in isolation and blurring the background. So, the image has to be masked to only keep the subject and remove any background. While the background many not be very useful, features detected in the background will only consume computing power and space.

Number of views #

For a cultural heritage model, it is important to take pictures of all possible angles and elevations as well as occluded regions so that they can be modeled. If a part of the structure is not present in at least 3 images and then the triangulation fails and the final output model may not contain the part. While this can be filled automatically in post-processing, the result may not be accurate.

Summary #

In summary, the following things needs to be considered.

  • Set the camera to use higher shutter speed, lower depth of field and lower ISO.
  • Close up pictures with more detectable features.
  • A 60-70% overlap of the detectable feature in at least two images.
  • Avoid shiny, specular and transparent subjects.
  • Avoid blurry images.
  • Avoid subjects with fine and many lose parts like hair, grids and nets.
  • Mask the subject from the background
  • Rotate and zoom the camera such that the subject covers at least 70% of the image.
  • Use different elevations for covering occluded and shadow areas like rooftops.
  • A larger number of photographs never hurts, except for the requirement of computing power.

With that in mind, the data collection part of the project may be carried out and the with enough data in the hands, we can move the next section of actual photogrammetric reconstructions.