Camera traps are ideal for automated capturing and monitoring wildlife, but traditional equipment often relies on infrared sensors to activate recording. Even a short delay is enough for small and fast-moving animals to leave the frame before recording starts, often leading to empty recordings.
In this paper, researchers proposed a novel camera trap triggering system based solely on a camera sensor. Intended for affordable use by citizen scientists, the system can be deployed using a cheap, low-power Raspberry Pi mini-computer with a camera and minimal storage requirements for less than €100.
Relying on innovative detector algorithms based on machine learning, the DynAikonTrap system incorporates a sequence of filters that ultimately leads to the storage of animal images, while automatically discarding video with no animals. While motion detection is faster than animal detection, a buffer between the two filters allows for near real-time analysis.
Using a set of 330 animal video recordings derived from GBIF-mediated occurrences, annotated by frame to indicate animal presence, the authors trained and benchmarked various configurations of detection filters. The ideal filter configuration of DynAikonTrap was able to continuously analyse and record more than 6 hours of animal presence per day at very low power usage.
Sharing their source code freely and openly, the authors provided complete instructions on how to deploy DynAikonTrap on inexpensive hardware, making camera trapping easy and affordable for both researchers and citizen scientists alike.