Colonies of bats represent some of the largest aggregations of mammals known to humankind. Censusing is important for understanding the ecological and economic impact of these gregarious animals. We have developed video analysis techniques for detecting, tracking, and counting bats as they emerge in dense formations from their caves. Since bats fly at night, we used multiple thermal infrared cameras to record and stereoscopically reconstruct their flight paths. Tracking large numbers of bats in multiple views is challenging because data association must be performed not only across time, as in single-camera tracking, but also across camera views; this is a multidimensional assignment problem, which is NP-hard. We present several techniques for occlusion reasoning, 3D tracking, and analysis of group behavior.

We have been developing a computer-vision based system that precisely and individually tracks a large number of living cells in a phase-sensitive (phase-contrast and DIC) microscopy image sequence, and detects cell events, such as migration (translocation), mitosis (division), apoptosis (death), and differentiation. The output is a complete cell lineage of the whole cell population, together with description and statistics of cells' shape, appearance, and motion. We are trying to make such technologies publically available through the Internet, hoping that such a capability of high-throughput spatiotemporal analysis helps biologists and tissue engineers to understand and direct cell growth. The talk will present the techniques, the system, and applications that we have worked on to date.

This lecture starts with sampling natural phenomena such as bird flocks, insect swarms, fish schools, smiling human faces, waving flags etc. that give rise to the abstract notion of dynamic particle system and deforming surface. It will then discuss the scientific values and potential applications of studying such natural phenomena, and will explain why 3D tracking is key to quantitative investigation into them that may help discover the underlying rules and models governing their motion.

By discussing a selected collection of existing methods for multi-view 3D tracking particle system and deforming surface, the lecture tries to assist the audience in appreciating the challenges facing the task of measuring 3D complex motion, and in building up a repertoire of useful techniques and in mastering their advantages and pinpointing their shortcomings.

In summary, the major aim of the lecture is to help consolidating the audience's quest for knowledge of the fascinating complex motion patterns in Nature, and to assist the audience in acquainting themselves with the very useful state-of-the-art computer vision technology.

• Appearance-based vs. tracking-by-detection
• Single target tracking
  • Recursive filtering
  • Dynamic programming
• Multiple target tracking
  • Filtering and data association
  • Multi-frame formulations

• Detection and counting
• Crowd flow analysis
• Social force models
• Detecting small groups