Researchers have developed new AI model that can identify and count animals in their natural habitat

• The new system used over 3 million images for training.
• This AI system has a 99.3% accuracy.
  
• Over 50,000 humans volunteered for the project.

Artificial intelligence (AI) is all around us from word suggestions to flying planes. And now researchers have developed an AI model which can automatically identify, count and describe animal activity in their natural habitats.

The new system uses deep neural networks, loosely inspired by how animal brains perceive the world. Such systems require vast amounts of training data for accuracy, and the data must be correctly labelled (like if the image is a positive or negative). Necessary data for this project was obtained from Snapshot Serengeti, which is a science project that has installed a large number of motion-sensor cameras in Tanzania that collects millions of images of animals in their natural habitats, such as deers, lions, leopards, cheetahs and elephants.

The system was trained using more than 3.2 million images which were captured using these cameras. The images were hand labelled by more than 50,000 humans over several years.

The result is that now the system can automatically identify animals in images with an accuracy of more than 99.3% while a crowdsourced team of human volunteers have an accuracy rate of 96.6%.

These performance results won't just help us improve our ability to study and conserve wildlife but also would save more than eight years of human labelling effort for each additional 3 million images.

As of now, the system can identify 48 different animal species in an image and can also give the count of each species present. The system also can tell what the animals are doing like - if they are eating, sleeping or if babies are present etc.