Poaching is a pandemic, and eradication of species like the black rhino is imminent unless an innovation turns the tide. The ability to locate, count and track animals is critical for wildlife conservation.
ERIC JÜRGEN HAASE, DTU-SPACE
A challenge to wildlife conservation is monitoring thousands of animals that move fast over vast areas. Mapping animal populations and distribution from the air allows assets to be deployed more effectively to guard endangered species against poaching or prevent human-wildlife conflict.
Wildlife census, or game counting, is a global industry serving both commercial and conservation sectors. There are many threats to wildlife best addressed by aerial surveillance.
Poaching
Poaching is primarily for ivory, rhino horn or meat. In Africa an elephant is killed every 15 minutes and rhinos are being decimated. Many poachers are now well organized, have helicopters, machine guns, grenades and allies in government and the military. Park rangers operate at a disadvantage and need aerial surveillance to oppose the odds.
Habitat loss
Airborne surveys can rapidly identify regions of deforestation or settlement within protected areas. Cattle sharing waterholes with wildlife is a vector for exchange of hoof and mouth disease, tsetse flies, etc., and the destruction of habitat pushes wildlife and humans closer together.
Human-wildlife conflict
Wildlife are killed if deemed ‘problem animals’. In southern African terms a cow is worth 3 months’ salary, so if predators kill livestock they are shot or poisoned in retaliation. As a result lions, leopards, cheetahs and hyenas are in rapid decline.
Human deaths often occur trying to stop elephants from raiding crops. But hippos and crocodiles claim the most victims by far, and are a threat to anything in a river. Monitoring the positions of dangerous animals could help avoid tragedies, loss of property and ultimately save wildlife too.
Hunting
Trophy hunting is big business in Africa and animals are the capital. For example hunting a lion costs about 50,000 USD and a black rhino license recently auctioned for 300,000 USD. Land is appraised in part by a wildlife census but is also used for conservation to determine a species’ protection status and regulate legal hunting.
Wildlife census
Conventional game counting is often done by one or two aircrafts flying very low to frighten animals out of vegetative cover, allowing observers onboard to spot them. This method is expensive, coarsely accurate and stressful to the animals.
Aerial photography offers an alternative by flying at altitudes less disturbing to wildlife and also documenting the populations. Typically an aerial survey collects thousands of images which take days of tedious inspection to search for animals, but to effectively address poaching or conflicts image evaluation must be rapid. Surveys in Namibia had 2% of imagery containing big game so any method to isolate images with animals reduces workload greatly. The solution is digital image classification and the techniques employed depend upon the type of cameras used.
RGB color
Red, green and blue ‘true color’ cameras can produce high spatial resolution imagery enabling visual identification of animals, but at the altitudes practical for mapping discrimination of similar species is challenging using amateur cameras. Professional cameras are an investment but offer wide fields of view with high spatial detail and 16 bit radiance that is ideal for inspecting highlights and shadows present in natural scenes.
Shape recognition is often used to classify RGB imagery. However wildlife often shelter in vegetation obscuring their topology, so segmentation may miss them, while rocks and logs provide abundant false positives. Reliable classification of RGB imagery has proven limited.
Thermal
Thermal cameras portray mammals as bright pixels indicating an animal in the scene but little species information. Photography can augment interpretation but thermography functions best at night. Reptiles may have low or negative thermal contrast, small animals might not register and dead animals will be missed. However rocks or bare soil can be look-alikes in warm climates.
Spectral
Multi and hyperspectral cameras measure light in the ultraviolet, visible and infrared wavelengths to distinguish bulk chemical compositions of a scene. Every substance has a characteristic pattern of reflectance known as a spectral signature that facilitates its identification within an image. If a single pixel has the signature of an animal then its presence is established even if the shape is obscured. Dead animals retain (nearly) the same spectra so poached carcasses can be found too.
CCD/CMOS cameras are sensitive over the wavelength range that characterizes pigments like chlorophyll, carotene and melanin which lend coloration to plants and animals. Exploiting their spectral differences they can defeat camouflage of wildlife, clothing or vehicles to distinguish targets in the field which is a useful anti-poaching device.
Conclusions
Multispectral cameras with infrared can distinguish fur-covered animals from pachyderms or reptiles. Sub-dividing these taxa requires hyperspectra to detect subtle details. As the technology becomes more accessible spectral methods will play an important role in wildlife counting and protection.