This thesis describes the development and implementation of the BeesBook System, a computer vision based solution for the automatic detection and analysis of behavioral patterns of honey bee colonies at the individual and collective level.
The behavioral analysis of honey bee colonies requires extensive data sets describing the behavior of individual colony members. These data sets must often be created manually - a time consuming and cumbersome activity. Consequently, behavioral data sets are usually restricted to small subsets of the colony’s life, whether this regards to time, space or animal identity. By automating the data acquisition process, the BeesBook system allows the supervision of a higher number of individuals during more extended periods of time, opening the door to more sophisticated, inclusive and significant studies.
The BeesBook System uses unique binary markers attached to the bees to keep track of their position and identity via computer vision software. The markers’ flexible design allows the implementation of a diversity of error-correcting codes, depending on the study’s goals and the colony’s population size. The markers adapt to the bee’s thorax shape creating a surface that withstands heavy-duty activity in and outside of the hive.
Three recording seasons were conducted during the summers of 2014, 2015, and 2016 to evaluate and improve the performance of the system components. Each season extended over a period of nine weeks and generated approx. 65 million images.
Prior to the beginning of each season, all members of a bee colony were individually tagged and transferred to an observation hive. The activity inside the hive was recorded using an array of four high-resolution cameras and stored for later analysis on one of the complexes of the North-German Supercomputing Alliance. Communication dances were identified in real-time using a second set of cameras comprised of two webcams running at high frequency. During the off-season, the experimental design was optimized to ensure that the generated data better serve the target of the experiment.
Stored images were processed using highly optimized computer vision software to obtain the position, orientation, and ID of every marked bee. These data are further processed to generate motion paths for the colony members, which, combined with data on the communication dances, constitute an unprecedented set of knowledge on the inner life of the honey bee colony. The information obtained through this system establishes the conditions for consolidating our understanding of already known behaviors. Furthermore, this research has identified previously unknown behavioral
data which ultimately extend our knowledge of bee colonies and their collective intelligence.
PDF-Datei von FUDISS_thesis_000000106153
Falls Ihr Browser eine Datei nicht öffnen kann, die Datei zuerst herunterladen und dann öffnen.