Tracking large numbers of densely-arranged, interacting objects is challenging due to occlusions and the resulting complexity of possible trajectory combinations, as well as the sparsity of relevant, labeled datasets. Here we describe a novel technique of collective tracking in the model environment of a 2D honeybee hive in which sample colonies consist of N ~ 103 highly similar individuals, tightly packed, and in rapid, irregular motion. Such a system offers universal challenges for multi- object tracking, while being conveniently accessible for image recording. We first apply an accurate, segmentation-based object detection method to build initial short trajectory segments by matching object configurations based on class, position and orientation. We join these tracks into full single object trajectories by creating an object recognition model which is adaptively trained to recognize honeybee individuals through their visual appearance across multiple frames, an attribute we denote as pixel personality. Overall, we reconstruct ~46% of the trajectories in 5min recordings from two different hives and over 71% of the tracks for at least 2 min. We provide validated trajectories spanning 3,000 video frames of 876 unmarked moving bees in two distinct colonies in different locations and filmed with different pixel resolutions, which we expect to be useful in the further development of general-purpose tracking solutions. Footnotes * https://groups.oist.jp/bptu/honeybee-tracking-dataset...

Pixel personality for dense object tracking in a 2D honeybee hive

10.1101/549006