A Framework for Understanding How Humans Perceive the Depth of Moving Objects

项目摘要 大脑的主要功能之一是从一系列2D图像中表示世界的3D结构 投射到视网膜上。在观察者平移过程中， 不同的距离（运动视差，MP）提供了有效的深度信息，在没有双目线索。 然而，如果对象在世界中移动，则这使从MP计算深度复杂化，因为将 是与物体在世界中的运动相关的图像运动的附加分量。以前的实验 并且关于来自MP的深度感知的理论工作假设对象在世界上是静止的。我们 建议使用人类心理物理学和计算模型的结合，首次， 人类如何在自我运动中推断移动物体的深度。 我们考虑两种方式，大脑可能计算移动物体的深度从MP。首先，如果大脑 可以准确地将视网膜图像运动解析为与自运动和物体运动相关的分量， 可以从由自运动引起的图像运动的分量计算。在目标1中，我们测试了这一点 假设，要求受试者判断物体的深度符号（近与远），这些物体是移动的或静止的， 世界我们假设，受试者的深度判断会受到世界中物体运动的影响， 最近的研究表明，流解析并不完全准确。我们的初步数据支持这一假设。 其次，我们认为大脑可能无法准确地分离图像运动的成分造成的 因为在推断物体是否在世界上运动时存在不确定性。这导致 我们假设目标1中观察到的偏差可以通过将感知视为联合推理来解释 世界上物体的运动和深度在目标2中，我们通过要求受试者回答两个问题来测试这一假设。 问题：1）物体在世界上是运动的还是静止的？2)物体是比注视点更远还是更近 重点？我们假设深度估计将取决于受试者对世界中物体运动的信念， 并且还应该系统地依赖于深度线索的可靠性。我们的初步结果支持预测 因果推理方案，我们将比较数据的贝叶斯理想观察者模型的预测。 该奖学金将通过以下方式为候选人提供计算和系统神经科学方面的培训： 与研究导师、罗切斯特大学更广泛的神经科学界的互动，以及 正式的课程这项拟议中的研究与NEI的目标一致，即“了解大脑如何处理 视觉信息”（国家眼睛和视觉研究计划）。此外，从这项工作中获得的知识 可以帮助我们更好地了解影响深度知觉的各种神经和眼科疾病， 并协助开发人工视觉和导航系统。

Project Summary One of the brain’s major functions is to represent the 3D structure of the world from a sequence of 2D images projected onto the retinae. During observer translation, the relative image motion between stationary objects at different distances (motion parallax, MP) provides potent depth information, in the absence of binocular cues. However, if an object is moving in the world, this complicates the computation of depth from MP since there will be an additional component of image motion related to the object’s motion in the world. Previous experimental and theoretical work on depth perception from MP has assumed the objects are stationary in the world. We propose to use a combination of human psychophysics and computational modelling to address, for the first time, how humans infer the depth of moving objects during self-motion. We consider two ways that the brain might compute the depth of moving objects from MP. First, if the brain can accurately parse retinal image motion into components related to self-motion and object motion, then depth can be computed from the component of image motion that is caused by self-motion. In Aim 1, we test this hypothesis by asking subjects to judge the depth sign (near vs. far) of objects that are moving or stationary in the world. We hypothesize that subjects’ depth judgements will be biased by object motion in the world, since recent studies suggest that flow parsing is not completely accurate. Our preliminary data support this hypothesis. Second, we consider that the brain may not be able to accurately isolate the component of image motion caused by self-motion because there is uncertainty in inferring whether or not objects are moving in the world. This leads us to hypothesize that the biases observed in Aim 1 can be explained by considering perception as joint inference of both depth and object motion in the world. In Aim 2, we test this hypothesis by asking subjects to answer two questions: 1) is the object moving or stationary in the world? 2) is the object farther or nearer than the fixation point? We hypothesize that depth estimates will depend on the subject’s belief about object motion in the world, and should also depend systematically on the reliability of depth cues. Our preliminary results support predictions of a causal inference scheme, and we will compare the data to predictions of a Bayesian ideal observer model. This fellowship will provide the candidate training in computational and systems neuroscience through interactions with the research mentor, the broader neuroscience community at the University of Rochester, and formal coursework. The proposed research is consistent with NEI goals to “understand how the brain processes visual information” (National Plan for Eye and Vision Research). In addition, the knowledge gained from this work may help us better understand the various neural and ophthalmological diseases that affect depth perception, and assist in the development of artificial vision and navigation systems.