Camera traps were originally used to confirm that a species was present at a site. The change that made them central to wildlife research was statistical rather than technical.
Detection is never complete
An animal can be present at a site and not photographed, because it passed outside the sensor's range, moved at the wrong angle or did not visit during the survey.
Raw counts of photographs therefore measure detection effort as much as animal abundance, and comparing sites on raw counts compares camera placement.
Occupancy modelling addresses this by using repeated visits to estimate the probability of detecting the species when it is present, and correcting the counts accordingly.
Individually marked animals allow direct estimation
Species with distinctive natural markings, such as tigers, leopards and many small cats, can be identified individually from photographs.
This turns a camera array into a capture-recapture study, in which the rate at which known individuals reappear indicates how many were never photographed at all.
Paired cameras facing each other are standard for this reason, since both flanks are needed to match an individual reliably across separate detections.
Unmarked species required a different approach
Most species cannot be told apart individually, which historically excluded them from abundance estimation by camera.
Newer methods use the rate of encounters combined with estimates of animal movement speed and the camera's detection zone to derive density directly.
These require additional information about behaviour and detection geometry, so they carry wider uncertainty, but they extend the method to a far larger set of species.
Array design determines what the data can support
Camera spacing has to relate to the target species' home range size, because cameras too far apart miss individuals and cameras too close sample the same animals repeatedly.
Placement introduces bias in a predictable direction, since setting cameras on trails and roads raises detection of species that use them and lowers it for those that avoid them.
Survey duration must be long enough to accumulate detections but short enough that the population can be treated as unchanged, which sets a practical window of weeks.
Image volume became the limiting factor
A moderate array generates enormous numbers of images, a large share of which are triggered by vegetation movement rather than animals.
Automated classification now handles the initial sorting and species identification, which has removed the bottleneck that previously limited study size.
Verification of a sample remains necessary, because classification errors are not random and can systematically affect the species that are hardest to identify.