Skip to main content

Robust Real-Time Object Detection

Citations: 1,648 | Paper Published: 01 Jan 2001 | Institutions: Mitsubishi Electric Research Laboratories | Authors: Paul A. Viola

Abstract in 2 Lines

This paper presents a framework for object detection that is fast and accurate. It introduces a new image representation called the "Integral Image", a learning algorithm based on AdaBoost, and a method for combining classifiers in a "cascade".

Contributions​

  • Introduction of a new image representation called the "Integral Image" which allows for very fast feature evaluation.
  • A learning algorithm based on AdaBoost which selects a small number of critical visual features and yields extremely efficient classifiers.
  • A method for combining classifiers in a "cascade" which allows background regions of the image to be quickly discarded while spending more computation on promising object-like regions.

Practical Implications​

  • The framework presented in this paper can be used for fast and accurate object detection in various applications such as face detection, pedestrian detection, and more.
  • The use of the Integral Image representation and the AdaBoost learning algorithm can significantly reduce the computation time required for object detection.
  • The method for combining classifiers in a "cascade" can further improve the efficiency of object detection by quickly discarding background regions of the image while spending more computation on promising object-like regions.
  • The insights and techniques presented in this paper can have broader applications in computer vision and image processing.

Methods​

Methodology
  • Integral Image representation: A new image representation that allows for very fast feature evaluation.
  • AdaBoost learning algorithm: A machine learning algorithm that selects a small number of critical visual features and yields extremely efficient classifiers.
  • Cascade method: A method for combining classifiers that allows background regions of the image to be quickly discarded while spending more computation on promising object-like regions.
  • Experiments on face detection: The proposed framework is demonstrated on the task of face detection and compared with the best previous systems.

Data​

The paper uses a face training set consisting of 4916 hand-labeled faces scaled and aligned to a base resolution of 24 by 24 pixels. The faces were extracted from images downloaded during a random crawl of the world wide web.

Results​

Performance Analysis

The paper presents a set of experiments on face detection using the proposed framework. The results show that the system yields face detection performance comparable to the best previous systems. The face detector constructed from 200 features yields a false positive rate of 1 in 14084 on a testing dataset, given a detection rate of 95%. The initial rectangle features selected by AdaBoost are meaningful and easily interpreted. The first feature selected seems to focus on the property that the region of the eyes is often darker than the region of the nose and cheeks. The face detection system implemented on a conventional desktop can process images at 15 frames per second.

Limitations​

  • The proposed framework is designed specifically for face detection and may not generalize well to other object detection tasks.
  • The training process for the detector is computationally expensive and time-consuming, taking weeks to train on a single machine.

Table of Contents