Recognition of indoor scenes using learning

点击翻译按钮获取中文摘要

The goal of this research is the development of improved vision-based object recognition methods for indoor scenes. Our research has focused on the following five major topics :(a) Recognition of indoor scene objects from color image sequences using human actions as a primary cue. We developed a system that first recognizes persons' actions ocurring in the scene and then, based on these, recognizes the objects which are manipulated by these persons(b) Recognition of indoor objects based on finger-pointing actions. Finger-pointing actions are detected and used to determine objects in the scene(c) Identification of persons based on human faces in unconstrained indoor environments. Development of a flexible face recognition system, based on a neural network bank(d) Recognition methods for autonomous driving of wheelchairs in outdoor environments(e) Fundamental techniques needed for improving sensor singnals and low-level data measurements in the context of the recognition problemi. Rotational imaging for the reconstruction of 3-D data points of scenesii. Background-frame differencing methods based on color similarityiii. Low-noise image acquistion using contemporary image sensor technology