[论文解读] Millisecond insect tracking system

T. Thang Vo‐Doan, Andrew Straw|arXiv (Cornell University)|Feb 27, 2020

Neurobiology and Insect Physiology Research参考文献 20被引用 8

一句话总结

该论文提出了一种毫秒级精度的昆虫追踪系统，利用高速云台镜面主动调节长焦摄像机与四象限光电二极管的光路，锁定昆虫身上25毫克的角反射器。系统实现了亚毫秒级闭环延迟，可在3米轴向范围及±40°水平与垂直视场内，实现高倍率、低运动模糊的视频录制，突破了传统视频在空间分辨率与追踪体积之间的权衡。

ABSTRACT

Animals such as insects have provided a rich source of inspiration for designing robots. For example, animals navigate to goals via efficient coordination of individual motor actions, and demonstrate natural solutions to problems also faced by engineers. Recording individual body part positions during large scale movement would therefore be useful. Such multi-scale observations, however, are challenging. With video, for example, there is typically a trade-off between the volume over which an animal can be recorded and spatial resolution within the volume. Even with high pixel-count cameras, motion blur can be a challenge when using available light. Here we present a new approach for tracking animals, such as insects, with an optical system that bypasses this tradeoff by actively pointing a telephoto video camera at the animal. This system is based around high-speed pan-tilt mirrors which steer an optical path shared by a quadrant photodiode and a high-resolution, high-speed telephoto video recording system. The mirror is directed to lock on to the image of a 25-milligram retroreflector worn by the animal. This system allows high-magnification videography with reduced motion blur over a large tracking volume. With our prototype, we obtained millisecond order closed-loop latency and recorded videos of flying insects in a tracking volume extending to an axial distance of 3 meters and horizontally and vertically by 40 degrees. The system offers increased capabilities compared to other video recording solutions and may be useful for the study of animal behavior and the design of bio-inspired robots.

研究动机与目标

• 克服高速动物视频录制中空间分辨率与追踪体积之间的权衡。
• 实现在大范围飞行过程中对昆虫肢体运动的高保真、多尺度观测。
• 通过主动光学追踪减少快速移动昆虫视频录制中的运动模糊。
• 实现低延迟闭环控制，实现实时追踪小型敏捷动物。
• 通过提供高分辨率、长时间的自由移动昆虫视频，支持详细的运动行为分析与仿生机器人研究。

提出的方法

• 高速云台镜面系统主动调节共享光路，在高分辨率、高速长焦摄像机与四象限光电二极管之间切换。
• 系统通过光电二极管的反馈锁定附着在昆虫身上的25毫克角反射器，保持精确对准。
• 光路实时动态调整，以毫秒级延迟追踪昆虫位置。
• 摄像机以高帧率运行，得益于主动指向机构，即使在高速下也能有效减少运动模糊。
• 追踪体积可达3米轴向范围，水平与垂直方向各±40度，支持大范围飞行观测。
• 系统将视频录制与位置反馈集成于单一闭环控制架构中，实现实时追踪。

实验结果

研究问题

• RQ1 主动相机转向能否显著减少高速飞行昆虫视频录制中的运动模糊？
• RQ2 闭环光学追踪系统在大三维空间内能将高空间分辨率维持到何种程度？
• RQ3 将四象限光电二极管与高速摄像机集成，如何提升追踪延迟与精度？
• RQ4 基于角反射器的主动光学系统在小型动物上可实现的最大追踪体积与分辨率是多少？
• RQ5 该系统能否实现可靠、长时间的复杂昆虫飞行行为观测，且运动模糊极小？

主要发现

• 系统实现了毫秒级闭环延迟，支持对快速移动昆虫的实时追踪。
• 由于主动相机转向，运动模糊显著减少，即使在昆虫高速飞行时亦然。
• 追踪体积扩展至3米轴向范围，水平与垂直方向各±40度，支持大范围飞行观测。
• 在整个追踪体积内成功录制了高分辨率视频，且运动模糊极小。
• 将四象限光电二极管与高速长焦摄像机集成，实现了对追踪系统精确的实时位置反馈。
• 原型验证了在飞行过程中对昆虫肢体运动进行多尺度、高保真观测的可行性。