260
Capturing Images of Flocks of Birds in Flight
Most of the case studies in this chapter focus on flocks of birds in flight: this is
because these present the greatest challenge to a bird counter. Often, however, flying
birds also offer the best opportunity of obtaining an accurate count of how many
birds are present. This is particularly true of gregarious estuarine waders such as
Dunlin and Knot, and of small and sociable ground-feeding birds, such as sparrows,
finches, larks and buntings.
As with feeding birds, birds in flight also like their own space: this would be
obvious in a 3D image but is much less evident in 2D images, where the birds will
overlap. This presents a problem for any automatic counter program that simply
counts the number of discrete objects in an image. There are ways to minimise the
error caused by overlapping birds in an image and these are covered both in the
section on validation below and in the individual case studies. However, life is
generally a lot easier if we can minimise the number of overlapping birds in images
at the time of capture. We can do this by following a few general rules:
(a) Identify the best vantage point to take the images, for example, as Starlings
arrive to roost or as waders arrive to feed on a falling tide;
(b) Try to capture the entire flock in a single image or, if the flock breaks up, then
take separate images of the component parts of it: large flocks of Starlings or
waders, for example, often aggregate and disaggregate during aerial displays;
(c) If the birds are in a tight flock, try to get an image of the flock as it banks to one
side or the other, this often reveals the spaces between the birds and reduces the
amount of overlap in the image; and
(d) If birds tend to arrive in sequence, e.g. Cranes or Geese, have a system for
recognising when one flock ends and the next one starts – perhaps by taking an
image of a tree or a building, something obvious that is in your field of view, to
form a break between the end of one flock and the start of the next.
Carrying Out Counts in ImageJ
There are two basic types of count available in ImageJ: a manual count and an automatic count. In general, a manual count is by far the best option if the image contains less than c. 3000 birds, and might be the only option if the image has a noisy
or complex background. Alternatively, if an image contains several thousands of
birds set against a relatively plain background, e.g. sky, water, mud or sand, then an
automatic count may be the more practical option. Both count types offer a more
reliable alternative to ‘by eye’ counts in the field.
C. Hurford
Capturing Images of Flocks of Birds in Flight
Most of the case studies in this chapter focus on flocks of birds in flight: this is
because these present the greatest challenge to a bird counter. Often, however, flying
birds also offer the best opportunity of obtaining an accurate count of how many
birds are present. This is particularly true of gregarious estuarine waders such as
Dunlin and Knot, and of small and sociable ground-feeding birds, such as sparrows,
finches, larks and buntings.
As with feeding birds, birds in flight also like their own space: this would be
obvious in a 3D image but is much less evident in 2D images, where the birds will
overlap. This presents a problem for any automatic counter program that simply
counts the number of discrete objects in an image. There are ways to minimise the
error caused by overlapping birds in an image and these are covered both in the
section on validation below and in the individual case studies. However, life is
generally a lot easier if we can minimise the number of overlapping birds in images
at the time of capture. We can do this by following a few general rules:
(a) Identify the best vantage point to take the images, for example, as Starlings
arrive to roost or as waders arrive to feed on a falling tide;
(b) Try to capture the entire flock in a single image or, if the flock breaks up, then
take separate images of the component parts of it: large flocks of Starlings or
waders, for example, often aggregate and disaggregate during aerial displays;
(c) If the birds are in a tight flock, try to get an image of the flock as it banks to one
side or the other, this often reveals the spaces between the birds and reduces the
amount of overlap in the image; and
(d) If birds tend to arrive in sequence, e.g. Cranes or Geese, have a system for
recognising when one flock ends and the next one starts – perhaps by taking an
image of a tree or a building, something obvious that is in your field of view, to
form a break between the end of one flock and the start of the next.
Carrying Out Counts in ImageJ
There are two basic types of count available in ImageJ: a manual count and an automatic count. In general, a manual count is by far the best option if the image contains less than c. 3000 birds, and might be the only option if the image has a noisy
or complex background. Alternatively, if an image contains several thousands of
birds set against a relatively plain background, e.g. sky, water, mud or sand, then an
automatic count may be the more practical option. Both count types offer a more
reliable alternative to ‘by eye’ counts in the field.
C. Hurford