Figure/Ground Segregation from Human Cues
Abstract
This paper presents a new embodied approach for segmentation by a humanoid robot. It relies on interactions with a human teacher that drives the robot through the process of segmenting objects from arbitrarily complex, nonstatic images. By exploiting movements with a strong periodic or discontinuity content, the robot's visual system segments a wide variety of objects from images, with varying conditions of luminosity and a different number of moving artifacts in the scene. The detection is carried out at different time scales for a better compromise between frequency and spatial resolution. The techniques presented ca be used in a passive vision system with a human instructor guiding the segmentation process. But a robot also may guide the process by itself, such as by poking or grabbing. The authors proposed a grouping strategy to segment objects that are not allowed to move and therefore may be difficult to separate from the background. This human-centered technique is especially powerful for segmenting fixed or heavy objects in a scene or to teach a robot segmenting through the use of books. The paper focuses on segmenting objects with similar color or texture as background, multiple moving objects in a scene, and objects in scenes that vary in robustness and luminosity.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2004
- Accession Number
- ADA434690
Entities
People
- Artur M. Arsenio
Organizations
- Massachusetts Institute of Technology