The VnBeeTracking Dataset
The VnBeeTracking Dataset is built for the purpose of serving the tracking honeybee problem.
The videos in the dataset are collected at the beehive entrance at the Research Center for Tropical Bees and Beekeeping, Vietnam National University of Agriculture in April 2022.
We used a data acquisition system consisting of an Nvidia jetson nano development kit and an IMX477 HQ camera with a 6mm CS-Mount lens which are placed in a housing surveillance weatherproof outdoor camera box. The camera is adjusted to the downward view and the camera box is attached to the upper area of the honeybee hive box as shown in the figure below:
Example of video collected by our data acquisition system:
Details
In order to capture images of honey bees at different times with different lighting conditions and numbers of honey bees, we recorded video over several days, each day between 6 am and 5:30 pm, each time is separate to 30 minutes, and each video lasts from 1 to 5 minutes.
5 videos recorded on mornings of different days were selected to build the dataset. These videos are recorded in 1080x720 resolution at 60 fps. We extracted frames from the first 10 seconds of each video, so the total number of frames extracted from all 5 videos is 3,000 frames. After that, for labeling all honeybees in these frames, we used the labelImg tool to determine the bounding box and assign ID for them. If a honeybee appears in multiple frames, the IDs assigned to it in these frames are the same. As a result, 252 IDs and 28,223 bounding boxes were annotated.
An example of a frame with labeled honeybees
The detailed statistics of VnBeeTracking dataset are shown in the table below:
| Video | Fs | IDs | BBs | Max IDs | Min IDs | Avg IDs | Max Fs | AVG Fs |
|---|---|---|---|---|---|---|---|---|
| Video1 | 600 | 25 | 5,416 | 11 | 7 | 9 | 600 | 217 |
| Video2 | 600 | 48 | 2,399 | 8 | 1 | 4 | 129 | 50 |
| Video3 | 600 | 44 | 2,955 | 9 | 2 | 5 | 599 | 67 |
| Video4 | 600 | 67 | 7,450 | 17 | 6 | 12 | 600 | 113 |
| Video5 | 600 | 68 | 10,003 | 24 | 12 | 17 | 600 | 147 |
where:
- Fs - number of frames in each video
- IDs - number of IDs in each video
- BBs - number of bounding boxes in each video
- Max_IDs - maximum number of IDs per frame
- Min_IDs - minimum number of IDs per frame
- Avg_IDs - average number of IDs per frame
- Max_Fs - maximum number of frames in which an ID appears
- Avg_Fs - average number of frames in which an ID appears.
Terms & Conditions of Use
The datasets are released for academic research only, and are free to researchers from educational or research institutes for non-commercial purposes.
Related Publications
All publications using VnBeeTracking or any of the derived datasets should cite the following papers:
- Le, T. N., Tran, D. N., Phan, T. T. H., Pham, H. T., Le, T. L., & Vu, H. (2023, October). A robust multiple honeybee tracking method from videos captured at beehive entrance. In 2023 International Conference on Multimedia Analysis and Pattern Recognition (MAPR) (pp. 1-6). IEEE.
Download
The dataset can be downloaded here. 5 original videos are placed in the folder named 'OrginalVideo'. The frames are extracted from the first 10s of each video and the .txt files that are the result of labeling the honeybees in the respective frames are placed in the folder named 'Image_Label'. The results of merging .txt files of each video are placed in the folder named 'GroundTruth' and you can use these files for evaluating the performance of your tracking model.