Automated detection of European wild mammal species in camera trap images with an existing and pre-trained computer vision model
Problem
Camera trap studies generate thousands of images that are expensive and time-consuming to analyze manually.
Why it matters
Manual image analysis limits the scale of wildlife monitoring for research and conservation.
What the AI did
A pre-trained FasterRCNN+InceptionResNet V2 model from TensorFlow Hub was tested on color and infrared images of ten European wild mammal species to detect animals and classify species.
Results
94% detection rate of correct region of interest; 71% classification accuracy for correct species name (abstract truncated mid-sentence).
Impact
Shows a pre-existing vision model can substantially automate camera trap analysis, though species-level accuracy is moderate.
Record
| Venue | European Journal of Wildlife Research |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Christin Carl, Fiona Schönfeld, Ingolf Profft, Alisa Klamm, Dirk Landgraf |
| Citations | 55 (OpenAlex, at import time) |
| References | 18 |
| Model or method | FasterRCNN+InceptionResNet V2 (TensorFlow Hub) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1007/s10344-020-01404-y |
| Open access | hybrid |
| Keywords | camera traps, deep learning, species detection, wildlife monitoring, FasterRCNN |
Importance rating
real but narrower advance
Abstract states each camera trap study generates thousands of pictures requiring expensive, time-consuming manual analysis, limiting wildlife monitoring scale; a real but narrow technical advance.