RUI: The Analysis of 3-D Motion for Visually-Guided Navigation

小行星9301326 当观察者在环境中移动时，眼睛表面会出现不断变化的视觉图像。 该图像中特征的运动传达了观察者在空间中运动的方向和速度的信息，以及环境中其他物体的三维运动信息。 已知仅从图像运动，人类观察者可以高精度地判断其相对于静止场景的3-D平移方向。 当环境中包含通过空间进行自身运动的物体时，从图像运动中恢复观察者的运动变得更具挑战性。 本研究将探讨人类视觉系统从变化的视觉图像中获得的信息来分析观察者和环境中物体的运动的机制。 心理物理学实验将研究人类在观看动态视觉显示时判断其3-D平移方向的准确性，动态视觉显示模拟观察者朝向包含移动物体的场景的运动。 进一步的实验将测试人类检测移动物体的能力，测量它们相对于观察者的三维平移方向，并判断观察者与接近物体碰撞的时间。 人类观察者通过复杂的动态场景的成功导航需要以高精度执行这些任务。 将开发计算模型来捕捉这些实验中观察到的行为。 这些模型将在计算机视觉系统中实现和测试，该系统从数字化的2-D图像序列中恢复移动的相机和移动物体的3-D运动。 这项工作的结果将进一步加深我们对人类系统如何使用图像运动分析来执行诸如在复杂场景中导航等任务的了解。 他们还将有助于开发成功的自主导航计算机视觉系统，以及需要从机器人、医学成像和监视等领域的动态图像中解释三维运动和结构的应用。 ***

9301326 HILDRETH When an observer moves through the environment, a continually changing visual image appears on the surface of the eye. The motion of features in this image conveys information about the direction and speed of movement of the observer through space, as well as about the three-dimensional motion of other objects in the environment. It is known that from image motion alone, human observers can judge their 3-D direction of translation relative to a stationary scene with high accuracy. The recovery of the observer's movement from image motion becomes far more challenging when the environment contains objects that undergo their own motion through space. This research will investigate the mechanisms by which the human visual system analyzes the motions of the observer and objects in the environment from information available in the changing visual image. Psychophysical experiments will examine the accuracy with which humans judge their 3-D direction of translation when viewing dynamic visual displays that simulate the motion of an observer toward a scene containing moving objects. Further experiments will test the human ability to detect moving objects, measure their 3-D direction of translation relative to the observer, and judge the time-to-collision of the observer with approaching objects. The successful navigation of human observers through complex dynamic scenes requires that these tasks be performed with high accuracy. Computational models will be developed that capture the behavior observed in these experiments. These models will be implemented and tested in a computer vision system that recovers the 3-D motion of a mobile camera and moving objects from a sequence of digitized 2-D images. The results of this work will further our knowledge of how the human system uses the analysis of image motion to perform tasks such as navigation through complex scenes. They will also contribute to the development of successful computer vision sys tems for autonomous navigation and to applications requiring the interpretation of 3-D motion and structure from dynamic imagery in domains such as robotics, medical imaging, and surveillance. ***