A System for Automatic Detection of Partially Occluded Objects from Real-World Images

Abstract

In this work we consider the Bayesian Integrate And Shift (BIAS) model for learning object categories and test its performance on learning and recognizing different object categories from real-world images. In contrast to conventional learning algorithms that require hundreds or thousands of training examples, we show that our system can learn a new object category from only a few examples. In addition, our system provides information not only about the object category but also about the local regions within the object on which it is fixating. We tested the performance of the system on very challenging examples of partially occluded targets. The training was done on different instances of one category and tested on partially occluded examples that the system had never seen before. We demonstrate that the system is very robust to partial occlusions and clutter and can recognize a target even if it fixates on the occluded part.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Nov 01, 2006
Accession Number
ADA481407

Entities

People

  • Leon Cooper
  • Liang Wu
  • Predrag Neskovic

Organizations

  • Brown University

Tags

Communities of Interest

  • Human Systems

DTIC Thesaurus Topics

  • Abstracts
  • Algorithms
  • Cognitive Workload
  • Computational Complexity
  • Computer Vision
  • Contrast
  • Detection
  • Detectors
  • Eye Movements
  • Learning
  • Object Recognition
  • Perception
  • Probability
  • Recognition
  • Training
  • Two Dimensional
  • Vascular System Injuries

Fields of Study

  • Computer science

Readers

  • Computer Vision.
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks