Soft and even lighting (no shadows)
Try to record your images with lighting that doesn’t create shadows in the scene or on the subject. If you are outside, a bright overcast day would be your best choice. If you are inside, try to create a lighting situation without shadows in or on your subject.
If you are recording images of large objects that might move (like plants or trees), pick a day without any wind. Any movement in the object during the image recording process will decrease the accuracy of your final model. Pick a dry day, wet conditions can cause reflections which will create imperfections in your final model.
Try and keep the exposure in all of your images consistent. Automatic exposure setting can cause some images to be lighter or darker than others so I suggest you set your camera on the manual exposure setting and record all of your images with one setting. If the lighting changes during the image collection procedure, consider recording images at another time. If it is not possible to reschedule your shoot, try and make sure the important parts of your image are exposed well.
Record lots of images
Record more images than you think you need. As a rule of thumb, you want to include every part of your subject in at least 5 different images. Consider recording images at a 10-degree offset from each other, this means you will record 36 images around the circumference of the object. Record a similar rotation of images above and below the centre point of the image and some images of the top and bottom of the subject.
Fill the frame
Make sure the subject fills the frame of each shot. Having wasted space around your image decreases the quality of your final model. Try vertical, horizontal and angle orientation to fill the frame.
Lots of overlap (terrestrial scenes)
If you are recording images of a large area. Try and stand in one location and record enough images to cover the scene (with at least 50% overlap in images). Then take a few steps parallel to the object and repeat this process. Continue making your way along the area of interest until you have traveled the length of the study area.
If you are recording images with a drone make sure you have 80% endlap (direction of the flightline) and 80% sidelap (distance between each flightline). Normally aerial images are recorded vertically (nadir looking) but you can also record images with a forward-looking angle (off-nadir or oblique). These forward-looking images can help create better detail in the vertical components of your model. To accomplish this, tilt the camera down at a 45-degree angle and fly the same flight pattern you would normally fly, then add a second flight sequence perpendicular to the first flight. This creates more images but it might also create a better 3D model.
Clear and sharp images
Try and keep your images as sharp as possible. Make sure you have the subject in focus, keep the camera steady and make sure your lens is clean.
Lots of texture will help
The Photogrammetric process relies on the software being able to identify the same points in multiple images so it will have a hard time creating a model from smooth, transparent, reflective or single-colored surfaces without any texture. Watch out for these types of surfaces and consider adding texture to them before recording your images.
Reoccurring patterns can cause problems
Watch out for areas containing patterns that repeat themselves throughout the scene. The Photogrammetric process relies on the software being able to recognise patterns of textures and colors in different images so if you have a similar pattern occurring throughout the scene, the software might have a hard time orienting and aligning the individual images correctly.
Turn OFF image stabilization
The Photogrammetric process relies on the camera and lens geometry remaining consistent for all images. Any shift in this geometry will cause problems when the model is created. The camera or lens image stabilization process intentionally shifts this lens and camera geometry to compensate for camera movement which will create inaccurate results in your final model.
Record measurements or GPS locations
Add some type of scale (ruler) in the image if you can. This scale can be used during processing to create a model with real world dimensions. If you are recording images of a large area, consider including objects in your scene with known GPS coordinates. Make sure these objects are well distributed throughout the area (corners, edges, and center areas). Record the GPS coordinates of each camera station (location of the camera when the images were recorded). These coordinates can be used to help assign real world coordinates to your model.
Return to Paul Illsley's home page