Capturing photos for photogrammetry
Photo capturing equipment¶
Technically photos for photogrammetric purposes can be captured with any camera. But some things do work better than others. Photogrammetric algorithms work best with generally high quality imagery. High quality photos can be obtained from many types of imaging devices as longs the technique and specifications are up to the job. A few key specifications regarding the equipment are recommended:
Resolution: 12 megapixels is a good starting point. Any less than that introduces a little risk that not enough detail could be captured for good 3D reconstruction results.
Lens focal length: anything between 10 and 135 millimeters of 35 mm equivalency should be good for photogrammetry, sweet spot being 14 - 50 mm equivalent lenses. All manufacturers specify the equivalent lens focal length because it is the most common unit to determine how wide or zoomed-in the lens is. Always refer to the cameras or lens' manual to find the equivalent focal length. Anything wider (less) than 16 millimeters is ultrawide angle which might introduce heavy distortion that might compromise reconstruction results. Anything longer or more 'tele' than 135 millimeters can result in shallow depth of field which also can make 3D reconstruction impossible. Special lenses such as fisheye, tilt-shift, soft-focus should be generally avoided due complicated geometric distortion introduced into images by using these lenses. Although, macro lenses are usually well suited for photogrammetry due to their excellent optical properties (almost no geometric distortions). Some lenses are advertised as having almost zero geometrical distortion, such lenses are well suited for photogrammetric purposes.
GPS geotagging: GPS data embedded to photos enables creating projects with scale and position which is required if any measurements are to be made. This is most applicable to drones. Almost all drones embed GPS data to photos because having a GPS sensor is an essential part of their operation. If the data is not embedded in the image exif data, flight tracks can be used. Flight track is a separate file containing the GPS data and Pixpro software can link photos to the data automatically. Flight tracks are usually used in advanced fixed wing UAV's. Some handheld cameras and most smartphones have GPS as well, but these are usually less reliable.
To achieve high accuracy positioning and measurement results, having GCP's or RTK level (survey grade) GPS data in the project imagery is recommended. Otherwise one can expect large absolute positioning errors (especially in Z axis) and somewhat less reliable measurements.
If these key specifications are met, with proper technique there should be no reason why one cannot capture images suited for high quality photogrammetric reconstruction results.
Photo capturing technique¶
Capturing suitable images is essential for photogrammetry. Suitable means that the images are nicely exposed and sharp at the same time. Acquiring such images is solely up to the user of imaging equipment:
Image sharpness: if the images are not sharp the 3D reconstruction will fail. Blurry images are usually a result of bad focus, motion blur, or extremely high ISO values. Bad focus occurs when a scanned object or terrain is missfocused. Some drones do have cameras which have fixed focus lenses which makes this a non issue, but those cameras that do have adjustable focus must be used with care. The rule of thumb is to always use manual focus. Set the focus on the scanned object or terrain and then turn the manual focus mode on, so that the focus will not change between images. If the distance between the object or terrain and the camera is extremely variable, first the flight plan should be adjusted and if that is not possible - autofocus can be used. Motion blur is another reason why 3D reconstructions fail. This is primarily applicable to drone imaging, because the drone is moving while taking photos. To avoid motion blur the camera must be stationary and stable or the images must be taken at a sufficient shutter speed. Shutter speed is a unit that defines how long is the cameras sensor exposed to light (hence another common synonymous term - exposure time). For example if the image is exposed for one tenth of a second (expressed as 1/10 in all camera systems) this is considered a slow shutter speed. So any camera movement while taking an image at 1/10 speed will be blurred to a degree. One should always try to maximize the shutter speed to make images free of motion blur. Distance to the terrain or object and speed of the drone are contributing factors, but as a rule of thumb do not go any slower than 1/100 shutter speed at any times when the aircraft is moving. Third reason why images can be blurry is extremely high ISO sensitivity values. ISO is a unit that determines how sensitive to light the camera sensor is. The higher the number - the more camera will amplify the signal that makes the image bright. The smaller the value - the better the image quality will be obtained, but if there is not enough light in the scene that you are capturing, the result will be a too dark or even a black picture. Balancing the shutter speed and ISO is the key to good aerial imagery. If the shutter speed is too low you will get motion blur, if the sensitivity is too high - you will get blurry images cause of noise. This is not usually a problem in well-lit situations such as a stockpile during the day, but if you feel that there is not a lot of light always remember to try to make sure that the balance between ISO and shutter speed is there without any motion blur or focusing issues.
Image brightness: almost black or white images will result in a failed reconstruction. Sharpness will not mean anything if you cannot actually see anything besides white or black in the image. Images made for photogrammetry, and indeed for anything else, should look nice. If you can see the features in the image, the software will see them as well. Balancing the previously mentioned settings as well as aperture is key to properly exposed images. Aperture is the final factor that makes up the photographs exposure. For photogrammetric purposes it is usually best to keep the aperture wide open. Wide open means that the lens itself will let through all the light it gathers. Aperture is expressed as an f number. The smaller the f number - the more open the aperture is. For example aperture value of f2.8 is a good small number that lets a large amount of light to the sensor. F11 on the other hand is a more closed aperture that will result in less light. Some drone have fixed aperture lenses which means that you do not have to worry about changing it whatsoever.
General photo capturing tips¶
To acquire the optimal quality of a 3D reconstruction, please follow the recommendations for capturing images:
- The overlap between the 2 consecutive photos should be at least 75 percent side ant 75 percent front.
- The distance from camera to object of interest should be as constant as possible to ensure consistent GSD.
The camera position angle varies respect to the type of target scanned subject. The optimal result is acquired with angles:
a. 90 degrees (nadir imagery or top-down photos), suitable for terrains and piles, orthophoto creation;
b. 45 degrees for the objects that very steep or perpendicular to the ground such as buildings;
For the objects of more complex geometry or homogenous, flat surface:
a. increase the overlap between photos up to 90 percent;
b. increase the camera altitude;
c. set camera exposure settings as accurately as possible so not to blow out any highlight or shadow detail.
Large and tall constructions (such as buildings, towers, monuments, etc.) should be captured in few stages:
a. 45 degree angled flight with consistent overlap.
b. 90 degree (nadir) flight with consistent overlap.
c. orbital flights around objects that require more detail or are very high.
In case of presence of a uniform or reflective surface on the terrain, increase the altitude and ensure capturing as many textured features as possible. Water, snow, sand, tin roofs and other uniform or reflective surfaces have too little keypoints for matching, thus they are almost impossible to reconstruct.
Plan the flight in respect with the terrain:
a. It is recommended to avoid capturing the sky, i.e. the photos should preferably contain the view of the terrain.
b. To avoid the presence of holes on the reconstructed object, make sure that the texture of the surface is rough enough. The view of the homogenous reflective surface (water, tin roofs, smooth road, snow, sand, plain flattened soil, etc.) might result in poor quality of reconstruction of the mentioned objects.
c. If multiple flights are to be made, ensure the consistent overlap between adjacent picture rows.
d. Building facades should be captured at 45 degrees or less if needed and the overlap should be approx. 90%.
Below and example of a proper simple nadir scan flight track with 80 percent overlap.
For elongated terrain scans, such as roads, railway tracks, river banks or just long and thin areas, special care should be taken to avoid the "bowl" or "bowing" efect. This effect occurs when only nadir imagery is used for elongated and relatively uniform objects. This effect will throw off any measurement accuracy and realistic representation of the scanned scene. To avoid this effect:
a. Use ground control. High accuracy ground control points can eliminate this effect.
b. Use RTK or PPK enabled imagery. High precision GPS RTK or PPK information embedded in the imagery is effective in reducing the bowing effect.
c. Add oblique imagery to the nadir images. Depending on the situation, adding oblique imagery in to the mix will reduce the bowing effect.
d. Increase the scanned area. Making the scan less elongated may help to reduce the bowing effect if more, non uniform features are added into the scan.
e. Use custom calibration data. If you have precisely calibrated lens' intrinsic values of the camera used for scanning, entering these values before reconstruction may reduce the bowing effect.
f. Make sure that image quality is as good as possible. Sometimes the bowing effect occurs if image quality is subpar.