The open source community is working together to provide ecologists and wildlife biologist with a time and cost-saving solution to the tedious work of sorting and classifying images captured by wildlife cameras. Their solution is MegaDetector, a free open source model that uses artificial intelligence to detect animals, people and vehicles in camera trap images.
Camera traps are a common method for monitoring wildlife across Canada, and across the globe. They offer a non-invasive way to gather data on animal presence and behaviour. However, a single study can collect thousands to hundreds of thousands of images, many of which are not of interest to researchers. Desired images have to be sorted and classified into groups, while empty images or those not of interest need to be removed. This is often a time intensive and expensive process. The good news for researchers who are tight on time or without the resources needed to manually slog through thousands of photos, is that MegaDetector is an accessible, free and time-saving solution.
The program uses artificial intelligence to speed up initial data processing by detecting what's captured in an image and sorting potentially thousands of photos into four broad categories (animal, vehicle, human, and empty). These can then be further classified or validated manually by humans. Manually sorting images can take researchers weeks to month to complete, whereas MegaDetector can process up to 10,000 images per day on an regular laptop, and up to 250,000 on a more powerful computer.
MegaDetector is publicly available on Github, free to use, and the model's collaborators offer support and training data to get researches started. The CCLM recently created an infographic on MegaDetector to help spread the word about this innovative and accessible model, and interested individuals can contact the National Boreal Knowledge Consortium to connect with researchers who have already applied MegaDetector to their work.