Automated identification of animal species in camera trap images
Problem
Manually identifying animal species in the huge volumes of images generated by camera traps is too slow for biodiversity monitoring.
Why it matters
Biodiversity studies generate thousands or millions of pictures that must be identified efficiently to advance the field.
What the AI did
An automated pipeline classifies cropped camera trap images using improved sparse coding spatial pyramid matching with dense SIFT and cLBP features, max pooling, and a linear SVM.
Results
Tested on over 7,000 camera trap images of 18 species from two field sites; accuracy not stated in abstract
Impact
Offers a way to scale species identification, with 231 citations suggesting notable field interest.
Record
| Venue | EURASIP Journal on Image and Video Processing |
|---|---|
| Published | 2013 |
| Work type | article |
| Authors | Xiaoyuan Yu, Jiangping Wang, Roland Kays, Patrick A. Jansen, Tianjiang Wang, Thomas Huang |
| Citations | 231 (OpenAlex, at import time) |
| References | 26 |
| Model or method | Improved sparse coding spatial pyramid matching (ScSPM) with linear SVM |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1186/1687-5281-2013-52 |
| Open access | gold |
| Keywords | camera traps, species identification, sparse coding, SIFT, SVM, biodiversity |
Importance rating
real but narrower advance
Abstract states biodiversity studies generate thousands or millions of pictures needing efficient identification; significant field-level challenge but not millions of lives.