Task-based ideal observer analysis and the efficiency of biological motion percep

描述(申请人提供)：人们能够很容易地将人类的运动与其他类型的运动区分开来。这种能力有时被称为生物运动感知，由于它在人类行为和社会互动中发挥的重要作用，它一直是许多科学兴趣的焦点。虽然已经进行了大量的研究来探索我们如何能够检测和识别人类运动的各个方面，但很少或根本没有研究试图量化人体运动携带了多少信息，或者人类观察者在试图检测、辨别或识别人类运动时能够使用多少信息。理想的观察者分析是一种能够定量回答这些问题的技术。我们的初步结果表明，与目前的教条相反，人类观察者对生物运动实验中通常使用的刺激和任务所携带的信息利用得很差。本方案中描述的第一组实验旨在通过使用基于信号检测的视觉信息处理模型和一系列系统识别技术来确定调节生物运动感知的机制，从而建立在这些结果的基础上。发现在生物运动感知任务中使用的信息如此之少，这让我们进入了第二个相关的研究领域。我们在生物运动感知任务中观察到的信息利用不足的情况，实际上在涉及复杂对象感知的任务中并不少见，例如面部和几何形状。这导致Use提出了这样一个问题：为什么在这些任务中信息处理如此糟糕？我们的假设是，在这些任务中信息使用效率低下的原因是由于心理物理实验中通常使用的高度人为和受限的条件。相反，人类的视觉系统已经进化成在刺激的可变性和不确定性的条件下是健壮的。在这项提案中，我们概述了一种新的方法，称为“基于任务的理想观察者分析”(TBIOA)，它涉及对任务的系统操作，以最大限度地提高人类利用信息的能力。这种方法的目标是让我们确定我们的知觉系统进化到最有效地处理哪些任务和刺激。在这个方案中，我们描述了一系列旨在实现和测试TBIOA的实验，包括简单和复杂的模式识别任务。除了上述两个研究方向的直接目标外，这项工作还有几个最终的应用。首先，这项研究可以用来为依赖于信息有效编码的应用程序的设计提供参考，例如视觉数据传输和压缩以及视频监控技术。其次，这项研究可能被用来帮助理解与视觉感知相关的残疾患者(例如视觉失认症)与正常观察者的不同之处，并告知如何设计人机界面来克服这种残疾。公共卫生相关性这项建议中描述的研究的第一个目标是了解当人们感知到他们的同胞的身体动作时，他们如何很好地使用他们可以获得的信息。第二个目标是开发和测试一种新的建模方法，称为基于任务的理想观察者分析，该方法寻求确定人类视觉系统最有效地处理哪些任务和刺激。这项研究对数据传输和压缩、视频监控技术的应用以及我们最终理解和治疗与视觉运动感知相关的疾病(例如，人们无法识别特定类别的视觉运动的视觉失认)具有潜在的重要意义。

DESCRIPTION (provided by applicant): People are able to easily discriminate the movement of humans from other kinds of movement. This ability is sometimes referred to as `biological motion perception', and it has been the focus of a great deal of scientific interest due to the important role it plays in human behavior and social interactions. Although a great deal of research has been carried out to explore various aspects of how we are able to detect and recognize human movement, there has been little or no research that has tried to quantify how much information is carried by the motion of the human body nor how much of this information a human observer is able to use when trying to detect, discrimination or identify human motion. Ideal observer analysis is a technique that allows one to quantitatively answer these questions. Our preliminary results have shown that, contrary to current dogma, human observers make very poor use of information carried by the stimuli and tasks typically used in biological motion experiments. The first set of experiments described in this proposal are designed to build on these results by using a signal-detection based model of visual information processing and a series of system identification techniques to specify the mechanisms that mediate the perception of biological motion. The finding that information use in biological motion perception tasks is so poor has led us to a second, related line of research. The poor use of information we have observed in biological motion perception tasks is actually not uncommon for tasks that involve the perception of complex objects, such as faces and geometric forms. This has led use to ask the question: why is information processing so poor in these tasks? It is our hypothesis that the cause of inefficient information use in these tasks is due to the highly artificial and constrained conditions typically used in psychophysical experiments. Rather, the human visual system has evolved to be robust under conditions of stimulus variability and uncertainty. In this proposal, we outline a new approach called `task-based ideal observer analysis' (TBIOA) that involves the systematic manipulation of a task in order to maximize the human ability to use information. The goal of this approach is to allow us to determine what tasks and stimuli our perceptual systems have evolved to deal with most efficiently. In this proposal, we describe a series of experiments designed to implement and test TBIOA with both simple and complex pattern discrimination tasks. In addition to the immediate goals of the two lines of research described above, there are also several eventual applications of this work. First, this research can be used to inform the design of applications that rely on efficient encoding of information, such as visual data transmission and compression as well as video surveillance technology. Second, this research could potentially be used to help understand how patients with disabilities related to visual perception (e.g., visual agnosias) differ from normal observers and inform how one might design human-machine interfaces to overcome such disabilities. PUBLIC HEALTH RELEVANCE The first goal of the research described in this proposal is to understand how well people use the information that is available to them when they perceive the bodily motions of their fellow humans. The second goal is to develop and test a new modeling approach called "task-based ideal observer analysis", which seeks to determine what tasks and stimuli the human visual system deals with most efficiently. This research has potentially important implications for applications related to data transmission and compression, video surveillance technology and for our eventual understanding and treatment of disorders related to the perception of visual motion (for example, visual agnosias in which people are unable to recognize particular classes of visual motion).