The synthesis of manipulator control programs from task-level specifications.
The synthesis of manipulator control programs from task-level specifications.
Strategy construction using a synthesis of heuristic and decision-theoretic methods.
Strategy construction using a synthesis of heuristic and decision-theoretic methods.
Computer Vision and Image Understanding - Special issue: Attention and performance in computer vision
S-SEER: selective perception in a multimodal office activity recognition system
MLMI'04 Proceedings of the First international conference on Machine Learning for Multimodal Interaction
| Hi-index | 0.00 |
This paper is a condensed version of the author's thesis [Bolles 1976], which investigates a subclass of visual information processing referred to as verification vision (abbreviated VV). VV uses a model of a scene to locate objects of interest in a picture of the scene. The characteristics that distinguish VV from the other types of visual information processing are: (1) the system has a great deal of prior knowledge about the type, placement, and appearance of the objects that form the scene and (2) the goal is to verify and refine the location of one or more objects in the scene. VV includes a significant portion of the visual feedback tasks required within programmable assembly. For example, locating a screw hole and determining the relative displacement between a screw and the screw hole are both VV tasks. Two types of VV tasks are discussed in the thesis: inspection and location. This paper only discusses location tasks, but essentially the same capabilities are required for both types of tasks. This paper describes (1) a structure for a VV system that makes it easier for programmers who are not experts in computer vision to program VV feedback, and (2) a set of combination rules that are capable of using the results of several different types of operators to estimate the confidences and precisions that are necessary within VV. An interactive VV system based upon these ideas has been implemented. It helps the programmer select potentially useful operator/feature pairs, provides a training session to gather statistics on the behavior of the operators, automatically ranks the operator/feature pairs according to their expected contributions, and performs the desired task. The VV system has also been interfaced to the AL control system for the mechanical arms and has been tested on tasks that involve a combination of touch, force, and visual feedback.