Collaborative Research: Automating the Large-Scale Measurement of Insect Behavior

格鲁吉亚理工学院和亚利桑那州立大学获得了资助，以开发一种综合方法，从视频记录中自动测量昆虫行为。昆虫行为的研究在生物学中起着基础性的作用，但进展受到数据收集速度的限制。研究人员在很大程度上依赖于直接观察或耗时的手动注释视频记录。该项目将创建一个自动化的解决方案，结合理论，算法，软件模块和行为测量数据库。这些工具将广泛应用于动物行为的研究，但开发将集中在蚂蚁提供的特别丰富和具有挑战性的问题上，其中必须同时跟踪多个相互作用的动物。当前的多跟踪技术在处理这种情况下的巨大程度的目标相互作用（包括一个目标被另一个目标遮挡的显著时段）的能力方面是有限的。本项目将产生一种新的方法，将图切割优化方法应用于视频对象分割。这种方法将能够识别视频的哪些部分对应于不同的目标，即使它们重叠。准确的目标分割还将有助于更准确地适应由于照明或其他环境影响而引起的外观变化。该项目还将开发新的行为识别方法，从目标配置和外观推断行为。与传统方法不同，这种方法不依赖于跟踪器的状态，从而避免了识别错误与跟踪错误的复合。首先是专注于与模块化软件工具兼容的算法和方法，从而使生物学家能够为广泛的传感任务开发定制的解决方案。第二个主题是利用最先进的超高分辨率成像传感器，以获得更多的信息，蚂蚁的行为和身份比目前可能的。这些能力将使昆虫生物学家能够构建和回答超出当前方法有限数据收集能力的研究问题。算法和软件模块将被广泛传播，以最大限度地发挥其在更普遍的环境中改造生物学的能力。欲了解更多信息，请访问项目网站www.example.com

The Georgia Institute of Technology and Arizona State University are awarded grants to develop an integrated approach to automating measurements of insect behavior from video records. The study of insect behavior plays a fundamental role in biology, but progress is limited by the rate at which data can be gathered. Researchers have relied largely on direct observation or time-consuming manual annotation of video records. This project will create an automated solution that combines theory, algorithms, software modules, and databases of behavior measurements. These tools will be widely applicable to studies of animal behavior, but development will focus on the particularly rich and challenging problems offered by ants, where multiple interacting animals must be simultaneously tracked. Current multi-tracking technologies are limited in their ability to deal with the huge degree of target interaction in this context, including significant periods of occlusion of one target by another. This project will generate a novel approach that applies the graph-cut optimization method to video object segmentation. This method will be able to identify which portions of the video correspond to distinct targets even when they overlap. Accurate target segmentation will also facilitate more accurate adaptation to changes in appearance due to lighting or other environmental effects. The project will also develop novel behavior recognition methods that infer behavior from target configuration and appearance. Unlike traditional methods this approach will not rely on the state of the tracker and thus will avoid the compounding of recognition errors by tracking errors.Two cross-cutting themes inform this project. The first is a focus on algorithms and methods compatible with modular software tools, thus allowing biologists to develop a customized solution to a wide range of sensing tasks. The second theme is the utilization of state-of-the-art ultra-high resolution imaging sensors to obtain more information about ant behavior and identity than is currently possible. These capabilities will enable insect biologists to frame and answer research questions that exceed the limited data collection capabilities of current methods. Algorithms and software modules will be widely disseminated, to maximize their power to transform biology in a more general setting. For more information visit the project website at http://www.kinetrack.org/