Analyzing Honey Bee Flight with Event-Based Vision

Abstract

An estimate of bee hive activity allows beekeepers and researchers to better understand trends in a colony’s health. This work presents a system utilizing an event-based vision sensor (e.g., Dynamic Vision Sensor, or DVS) to track flying bees in real-time with the intent of accurately monitoring the flow of bees in and out of an Apis mellifera colony. Neuromorphic event-based vision sensors like the DVS are well-suited to the detection of small, fast-moving bees with minimal latency due to the asynchronous pixels. Rather than processing and transferring full images, these pixels detect changes in brightness independently, only sending updates where movement occurs, dramatically reducing the computational load. Using this spatio-temporal input, event-based algorithms are able to track fast-moving bees in real-time to determine the position of the bee relative to the hive entrance, and by defining a boundary, count the number of bees leaving and returning. Due to the sensor’s temporal resolution, the flapping bee wing can be observed in flight and its wingbeat frequency can be estimated during tracking in real-time. To evaluate the proposed event-based tracking system, a side-by-side comparison with a frame-based camera at an active colony was performed. Real-time tracking of trends in bee activity should provide early warning signs of problems such as robbing, swarming, absconding, etc. Detailed analysis of wingbeat frequency may eventually provide a real-time detection system for invading insects.