We were approached by Dr. Michael Smith to design a high resolution recording system to capture, continuously, the behavior of an entire honey bee colony for a season (approximately 4 months).
Michael explained his research goals;
The project we're undertaking looks at how honey bees change the way they interact with each other, and their environment, throughout their lives. To do this, we wish to record the behavior of an entire hive for an extended period of time. Being able to conduct long-term recordings of a honey bee colony allows us to take a behavior of interest, and then look at that individual's experience days, or even weeks before.
In collaboration with Prof. Dr. Tim Landgraf Michael will quantify behavior by affixing to each bee an individual 'beesbook' tag, and then tracking the honey bee's identity through the season.
The combination of continuous recording and long-term tracking of honey bee colonies allows us to pursue research questions that would be impossible in a traditional setup.
To meet the requirements we designed a Motif based solution using 4x 12 Megapixel cameras at 3 frames per second. To save on storage space, all video is compressed in real-time, at exceptionally high quality so it is compatible with tag tracking, before being saved to disk.
Because the Motif user-interface is web based and can be accessed remotely, the entire colony can be observed from the safety and security of ones office.
The honey bees live in a nest of beeswax combs, sandwiched between two glass sheets. Each side of the hive is imaged by two cameras. To provide sufficient light, a strobed lighting system is employed:
- Bright infrared lights and a pair of cameras on one side of the hive flash synchronously at 3Hz.
- Also at 3Hz, but shifted in time relative to the first two cameras, an identical arrangement of cameras and lighting observing the other side of the hive flash and capture an image.
At regular intervals Motif's built in support for copying data to network storage was used to safely backup the data to central infrastructure.
After both images are stitched together, the 'beesbook' algorithm is used to detect the honey bee identity.
Hover your mouse over the image to see a zoomed-in view of the tracked honey bees
The data-set collected is extremely impressive:
- 2 Seasons of Recording
- 80 days of recording per season, with 60 days to be used for analysis
- 16TB of video data per camera per season