The Arribada conservation tech initiative is working with ZSL Whipsnade Zoo to develop an early warning thermal camera system to help reduce human-elephant conflict in the wild. The project will use machine learning to automatically recognize the thermal photo of an elephant and send an alert to communities that an elephant has been spotted nearby. Alerts will reduce surprise elephant encounters, which are often the cause of elephants injuring humans and, in retaliation or fear, humans hurting elephants. Teaching a computer to automatically recognize an elephant in thermal vision requires thousands of pictures and the elephant habitat at Whipsnade Zoo is the perfect place to gather them. With the help of the keepers, Arribada has been collecting pictures over the last four months with special thermal cameras.
This information will help not only train a computer but test the limits of this technology. It's key to be able to answer a number of questions - for example, at what distance does the computer stop being able to identify the pattern of white pixels as an elephant? Is there an optimum temperature window, above which, elephant bodies become one with the scenery and identification becomes impossible? Answering these vital questions will help understand how these cameras will perform in the real-world, where communities will depend on them to help reduce conflict the face with elephants almost daily. Taking photographs of humans is also important - in order to prevent emergency elephant alerts being sent to communities every time a person walks past the cameras. The keepers have worked hard to assist with this project as much as they can.
Arribada and ZSL will continue to develop and test the thermal cameras with Whipsnade Zoo’s elephants before taking them into real communities. We hope this project can connect people to ZSL’s conservation projects and show how animals in zoos can play a vital role in the conservation of their wild counterparts, help preserve wild animal populations and better their livelihoods.
