The goal of this thesis is to present the first realtime honey bee waggle dance decoding system, and to establish a benchmark of detection rates and decoding quality. Honey bee research concerns a wide scientific community, but here are no automatic waggle dance detection and decoding systems in use. Hence the goal of this thesis is to create one.

This master's thesis first introduces the honey bee waggle dance communiction and motivates the need for an automatic decoding system. Then it examines related work and concludes that no such systems exist. First, I define the software requirements, then I present a detailed implementation. Finally, the presented algorithm is applied to video recordings and used in two experiment. Decoding quality is evaluated with precise video material. Detection results of six and four weeks live observations are evaluated. Decoder system's capability to reconstruct the location of a trained artificial food source is tested.

Evaluation results show that the realtime honey bee waggle dance decoding system can achieve 96.4% precision rate and 89.5% recall rate with recorded videos. The system shows 78.6% and 67.4% precision rates for live observations. On average decoded dance positions show 0mm error and a low standard deviation error. Waggle run duration decoding averagely exhibits an error of additional 97.79 ms and a standard deviation error of 138.52 ms. On average the decoding of honey bee dance directions shows -1.06° error and a standard deviation error of 35.83°. In the end I discuss different evaluation results and present further improvements for future works.