Single Review on past events

Many biologists are interested in studying the behavior of animals. Some of them are interested in bees, and among them are a few that are primarily specializing in the movements of bees´ antennae, proboscis and mandibles to get to know more about olfactory learning. That the analysis of these movements is not that easy seems quite obvious: The movements are fast and minute. Here computer expert Minmin Shen gets involved: She developed a computer program designed to analyze video data from bee experiments. In her talk about “An interactive framework for insect tracking” she presented her research.

Tracking is defined as “estimating location of moving object(s)/extract trajectory”. The video data she receives from biologists contain tracked bee movements in a rate of 30 or 60 frames per second and lasting 15 sec to 20 minutes. She classifies the tracked objects into three categories: antenna, mandibles, and proboscis. Then she estimates benchmark frames and creates a constraint frame-to-frame linking. Because of the interaction between targets tracking is quite difficult: detection errors occur, and missing, occluded or merged targets create a long tracking gap.

“Here lie the limitations of automatic approaches”, explains Minmin Shen. “The labels will not be corrected with the help of the benchmark frames. Even sophisticated algorithms have their limits, if the assumptions do not hold.” Hence evaluation of tracking performance has to be done manually by viewing through the whole video. Thus Minmin Shen proposes an interactive tracking algorithm. She requests users to correct a small portion of the whole video so that the frames with high annotation cost (key frames) are rectified.

In the future Minmin Shen will work on a more sophisticated mathematical model of the trajectories and track the targets at the precision of pixel level.