Tracking multiple cows simultaneously in barns using computer vision and deep learning
This study investigated the automated tracking of multiple cows simultaneously using computer vision and deep learning. Video clips were collected in 2019 at Dairy Campus, where cows were housed in small groups (n=16). A systematic approach covering the true variability of barn circumstances eventually resulted in the selection of 159 frames that were annotated by drawing bounding boxes around each cow. These frames were used to retrain and test four You Only Look Once version 5 (YOLOv5) models to automatically detect cows. The weights of the best performing YOLOv5 model were used to parametrize the deep learning algorithm DeepSORT to track multiple cows simultaneously. This algorithm was applied to a 10 min timeframe of a randomly selected video clip and evaluated by computing the multi-object tracking accuracy, which was 92.8%. This outcome is a promising and essential step towards automated monitoring of individual behaviour of group-housed cows.
| Main Authors: | , , , |
|---|---|
| Format: | Article in monograph or in proceedings biblioteca |
| Language: | English |
| Published: |
Wageningen Academic Publishers
|
| Subjects: | Life Science, |
| Online Access: | https://research.wur.nl/en/publications/tracking-multiple-cows-simultaneously-in-barns-using-computer-vis |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study investigated the automated tracking of multiple cows simultaneously using computer vision and deep learning. Video clips were collected in 2019 at Dairy Campus, where cows were housed in small groups (n=16). A systematic approach covering the true variability of barn circumstances eventually resulted in the selection of 159 frames that were annotated by drawing bounding boxes around each cow. These frames were used to retrain and test four You Only Look Once version 5 (YOLOv5) models to automatically detect cows. The weights of the best performing YOLOv5 model were used to parametrize the deep learning algorithm DeepSORT to track multiple cows simultaneously. This algorithm was applied to a 10 min timeframe of a randomly selected video clip and evaluated by computing the multi-object tracking accuracy, which was 92.8%. This outcome is a promising and essential step towards automated monitoring of individual behaviour of group-housed cows. |
|---|