AI cameras can spot elephants and alert researchers even with no cell phone signal

Researchers trained special AI cameras to recognise and photograph elephants in a Gabon forest, and to then alert the researchers that they spotted these large trunked animals. 

Although the cameras sometimes mistook branches for trunks, the technology’s main appeal is that it uses only a small amount of data to send alerts via satellite from anywhere in the world - a feat that could hugely aid conservation efforts.

People use automatic cameras in many ways, such as for wildlife tourism, ecosystem monitoring, and detecting illegal hunting of animals. Many of these uses need the camera to alert people as events occur. Unfortunately, cameras that can do this are uncommon.

Most places where people want to use these cameras also do not have reliable cellular or WiFi signals. This either leads to a long delay in sending an alert, or people needing to check the camera manually.

These researchers wanted to solve this problem by creating a camera that can send alerts to users anywhere in the world. In addition, they wanted the camera to be low-cost, usable in rural areas without existing cellular signals, and easy to use. They also wanted to create guidelines on how others could create similar or better cameras.

They created a camera that uses artificial intelligence (AI) and connects to the Iridium satellite network (ISN). The ISN is a group of machines that orbit around the earth and allow people to send information to each other.

Anyone can pay to send messages over the ISN from anywhere on Earth. But, it is cheaper to send smaller messages, for instance text rather than images.

So, instead of sending images from the camera, they taught an AI program how to sort photos into categories, ran the program on the camera, and had the camera send a prediction of the image's category instead of the whole photo. 

They tested 7 of these AI cameras for 72 days in Gabon, where they taught them to detect elephants. Whenever the camera took a photo of an elephant, it sent an alert to the researchers. This was to see if these cameras could help local farmers protect their crops from elephants.

Their cameras could take about 17 photos a day, for about 3 months, before they needed to replace the batteries. They received alerts about 7 minutes after the camera took the photo. 

With some adjustments, the camera improved from correctly detecting elephants 84% of the time, to 98% of the time.

The researchers hoped that making their designs and reasoning available for others to access (open access) would help improve their camera through further research. 

Other researchers had previously developed similar technology for detecting penguins and lions, but this study looking at elephants specifically had several unique, new challenges to solve.

For instance, the dense tree cover made it difficult to install the cameras where they could connect to the ISN and get enough sunlight to charge (using solar power). The temperature and humidity also interfered with some of the charging circuits.

The researchers said that some branches that looked like elephant trunks caused the camera to make mistakes as well.

They were able to fix some of these issues, but they suggested that future studies should improve the AI program, use microphones, and figure out how to send low-cost images along with text alerts.

These researchers were from Gabon, South Africa, the UK, the Netherlands, Poland, and the USA. They focused on a forest in Gabon that may be similar to other African forests, so they hope the technology can be applied elsewhere too.

Abstract

Efforts to preserve, protect, and restore ecosystems are hindered by long delays between data collection and analysis. Threats to ecosystems can go undetected for years or decades as a result. Real-time data can help solve this issue but significant technical barriers exist. For example, automated camera traps are widely used for ecosystem monitoring but it is challenging to transmit images for real-time analysis where there is no reliable cellular or WiFi connectivity. Here, we present our design for a camera trap with integrated artificial intelligence that can send real-time information from anywhere in the world to end-users.

We modified an off-the-shelf camera trap (Bushnell™) and customised existing open-source hardware to rapidly create a ‘smart’ camera trap system. Images captured by the camera trap are instantly labelled by an artificial intelligence model and an ‘alert’ containing the image label and other metadata is then delivered to the end-user within minutes over the Iridium satellite network. We present results from testing in the Netherlands, Europe, and from a pilot test in a closed-canopy forest in Gabon, Central Africa.

Results show the system can operate for a minimum of three months without intervention when capturing a median of 17.23 images per day. The median time-difference between image capture and receiving an alert was 7.35 minutes. We show that simple approaches such as excluding ‘uncertain’ labels and labelling consecutive series of images with the most frequent class (vote counting) can be used to improve accuracy and interpretation of alerts.

We anticipate significant developments in this field over the next five years and hope that the solutions presented here, and the lessons learned, can be used to inform future advances. New artificial intelligence models and the addition of other sensors such as microphones will expand the system’s potential for other, real-time use cases. Potential applications include, but are not limited to, wildlife tourism, real-time biodiversity monitoring, wild resource management and detecting illegal human activities in protected areas.

Disclaimer

This summary is a free resource intended to make African research and research that affects Africa, more accessible to non-expert global audiences. It was compiled by ScienceLink's team of professional African science communicators as part of the Masakhane MT: Decolonise Science project. ScienceLink has taken every precaution possible during the writing, editing, and fact-checking process to ensure that this summary is easy to read and understand, while accurately reporting on the facts presented in the original research paper. Note, however, that this summary has not been fact-checked or approved by the authors of the original research paper, so this summary should be used as a secondary resource. Therefore, before using, citing or republishing this summary, please verify the information presented with the original authors of the research paper, or email [email protected] for more information.