Improving Object Detection for Time-Lapse Imagery in Wildlife Monitoring
Author Information
Author(s): Marcus Jenkins, Kirsty A. Franklin, Malcolm A. C. Nicoll, Nik C. Cole, Kevin Ruhomaun, Vikash Tatayah, Michal Mackiewicz
Primary Institution: University of East Anglia
Hypothesis
Can the performance of an object detector in time-lapse imagery be improved by incorporating temporal features from prior frames?
Conclusion
The proposed method significantly enhances object detection accuracy by integrating temporal features, achieving a 24% improvement in mean average precision.
Supporting Evidence
- The method achieved a 24% improvement in mean average precision over the baseline object detector.
- Temporal features were integrated into the YOLOv7 architecture to enhance detection accuracy.
- The study utilized a large dataset of approximately 180,000 images for training and validation.
Takeaway
This study shows how using pictures taken over time can help computers better recognize animals in images, making it easier to monitor wildlife.
Methodology
The study utilized a camera-trap dataset of breeding tropical seabirds, applying a YOLOv7 object detection model enhanced with temporal features from prior frames.
Potential Biases
Potential bias in the dataset due to varying visibility and annotation quality across different camera traps.
Limitations
The dataset may not represent all wildlife scenarios, and the method's effectiveness could vary with different species or environments.
Participant Demographics
The study focused on breeding tropical seabirds, specifically the Pterodroma petrels on Round Island, Mauritius.
Statistical Information
P-Value
0.0001
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website