Fast and accurate direction estimation of moving pedestrians

ABSTRACT

This work proposes a simple yet efficient way to estimate pedestrians flow direction based on videos from still cameras. It does that by localizing the extremities of head and feet of silhouettes and fitting them to lines. As the previous in three-dimensional space of these lines are parallel, their intersection point is the vanishing point. Using the computed vanishing point and two internal camera parameters, the horizontal direction of moving pedestrian is determined. Our method competes for the state-of-the-art methods and achieves a high rate accuracy for direction classification.

KEYWORDS:

• Projective geometry
• vanishing point
• direction
• video sequence

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes on contributors

Amina Bensebaa received the master degree in Intelligent Informatics Systems from the University of Sciences and Technology Houari Boumediene, Algeria, in 2012. She is also a PhD student at the University of Sciences and Technology Houari Boumediene, Algeria, under Professor Larabi’s supervision. Her current research interests concern image and video analysis, content-based image retrieval and pattern recognition.

Slimane Larabi received his PhD in Computer Science from the National Institute Polytechnic of Toulouse, France, 1991. In January 1992, he joined the Computer Science Department of USTHB University in Algeria, where he is currently a Professor. He leads research in Computer Vision Group of the Laboratory of Artificial Intelligence Research. His work spans a range of topics in vision including: image description, human action recognition, head and body pose estimation and video analysis. He also proposed and leads the Master of Visual Computing in the same university and teaches several courses: data visualization, game design, multimedia systems, artificial intelligence and computer vision. He conducted many projects in different areas, especially in computer vision, augmented reality, and data visualization.

ORCID

Slimane Larabi http://orcid.org/0000-0001-8994-5980

Image notes

The left image used in figure 2 can be accessed through the following link: https://www.dezeen.com/2017/10/12/umbrellium-develops-interactive-road-crossing-that-only-appears-when-needed-technology/ and the right image has been taken from ref [2].

The image dataset used for figures 3 & 4 can be accessed through the following links https://towardsdatascience.com/ http://www.cvg.reading.ac.uk/PETS2006/data.html

The images in figure 8 (left) and figure 11 were taken from the dataset MGP01 which can be accessed through the following link: http://perso.usthb.dz/∼slarabi/MGP01.html and the right image of figure 8 can be accessed through the following link: https://cvlab.epfl.ch/data/data-pom-index-php/