Fast Human Detection
We have a developed a human detector algorithm that is as accurate as state-of-the-art detectors, while being about 70 times faster.
Background & Objective: The problem of human detection in still images has been extensively investigated within the computer vision community. It is difficult because of the large variations in human appearance due to changes in illumination, camera position, clothing and body pose. We have developed a real-time system for human detection that combines a Histogram-of-Gradient representation, for high accuracy, with Integral Histogram representation, for real-time performance.
Technical Discussion: Our work combines two leading approaches to object detection. One that uses Histogram-of-Gradients (HoG) to represent objects in a robust manner and an Integral image representation that allows for fast and efficient implementation. We use an AdaBoost training algorithm to learn a cascade of rejecters that are based on HoG of windows of different size and position to quickly reject image patches that do not contain humans. The method compares favorably with other leading techniques in terms of accuracy, while running up to 70 times faster than comparable systems.
Publications:
| Technical Reports: | |
| Pedestrian Detection Using Boosted Features over Many Frames | |
| Fast Human Detection Using a Cascade of Histograms of Oriented Gradients | |
| Ensemble Tracking | |
Technology Areas:
Imaging
Computer Vision
Modification Date: January 16, 2009