Task Oriented Vision
Task Oriented Vision
复制标题
任务导向的愿景
DOI：
10.1109/iros.1992.602084
复制
发表时间：
1992
期刊：
影响因子：
--
通讯作者：
M. Hebert
中科院分区：
文献类型：
--
作者：
K. Ikeuchi;M. Hebert
在本文中，我们介绍了一个系统的方法，定制感知模块的特定任务。在这种方法中，图像分割，对象表示和操纵模块的选择基于目标任务的约束条件和环境。我们的目标是根据约束条件生成最有效的感知模块集。这种方法是一个高层次的相当于更传统的低层次的主动控制的传感策略。我们通过两个示例系统来说明这种方法。在第一个系统中，任务是操纵自然物体。在第二个系统中，任务是从垃圾箱中挑选工业零件。我们展示了如何分解和分析这两个任务，以产生感知能力和任务要求之间的最佳匹配。然后，我们介绍了一个基于任务的视觉系统开发的一般框架。
In this paper, we introduce a systematic approach for tailoring perceptual modules to specific tasks. In this approach, modules for image segmentation, object representation, and manipulation are selected based on the constraints of the target task and on the environment. The goal is to generate the most efficient set of perceptual modules based on the constraints. This approach is a high-level equivalent to the more traditional low-level active control of sensing strategies. We illustrate this approach through two example systems. In the first system, the task is to manipulate natural objects. In the second system, the task is to pick industrial parts out of a bin. We show how the two tasks were decomposed and analyzed to yield the best match between perceptual capabilities and task requirements. Then, we introduce a general framework for task-based development of vision systems.