Fundamental issues in cognitive computer vision

本研究项目属于认知计算机视觉领域。认知计算机视觉的研究人员开发数学模型、算法和计算机程序来自动确定数字图像和视频中出现的物体和场景的类型。尽管这个难题的全面解决方案还需要几年的时间，但在对该技术的实际应用产生直接影响的不同方面，已经取得了定期的进展。本研究计划旨在通过解决与研究人员面临的关键问题相关的基本问题，即在大范围距离和视点上检测和识别物体的可能性，改进算法以覆盖适当的一般类型的物体，以及减少运行程序所需的时间，从而进一步做出贡献。在一些重要的问题上取得进展，导致鲁棒、高效和有效的自动认知计算机视觉方法。每当应用程序需要理解仅允许有限特定期望的上下文中的图像内容时，就需要这些内容。因此，这项研究计划将帮助实现在安全，交通，媒体，医疗保健和教育领域的大量不同应用，所有这些都具有很高的社会经济影响。图像理解方法通常试图重现人类的表现，而没有建模和估计潜在大脑过程的参数。因此，在开发和测试认知计算机视觉方法时，需要一种适当的、独特的方法。最近，我们定义了这样一种方法，作为我们努力改进认知计算机视觉方法的开发、记录和评估方式的一部分。七名研究生将参与这项研究项目。过去的学生在大学、研究实验室和高科技行业担任职务。

This research program is in the field of cognitive computer vision. Researchers in cognitive computer vision develop mathematical models, algorithms, and computer programs to automatically determine what types of objects and scenes appear in digital images and videos. Even though a completely general solution to this difficult problem is still years away, regular progress is made on different aspects that have an immediate impact on real-world applications of the technology. This research proposal intends to contribute further by addressing fundamental issues related to key questions facing the researchers, namely the possibility to detect and recognize objects over a large range of distances and viewpoints, the improvement of the algorithms to cover appropriately general types of objects, and the reduction of the time it takes to run the programs. Progress is to be made on a number of important issues leading to robust, efficient, and effective automatic cognitive computer vision methods. These are required whenever an application necessitates an understanding of the image contents within a context that allows only limited specific expectations. As such, this research program is to help actualize a large number of different applications in security, transport, media, health care, and education fields, all with high socioeconomic impact. Image understanding methods usually attempt to reproduce human performance without modeling and estimating parameters of the underlying brain processes. For this reason, a proper and distinct methodology is required and must be followed in developing and testing cognitive computer vision methods. Recently, we defined such a methodology as part of our effort to improve the way cognitive computer vision methods are developed, documented, and evaluated. Seven graduate students are to work on the research program. Past students are holding positions in universities, research laboratories, and high-tech industries.