Skip to Main content Skip to Navigation

Détection et suivi de personnes par vision omnidirectionnelle: approches 2D et 3D

Abstract : In order to realize applications allowing 3D pose estimation, we investigate the problem of 3D people detection and tracking in omnidirectional images sequences. This requires a stable and accurate monitoring of the person in a real environment. In order to achieve this, we will use a catadioptric camera composed of a spherical mirror and a perspective camera. This type of sensor is commonly used in computer vision and robotics. Its main advantage is its wide field of vision which allows it to acquire a 360-degree view of the scene with a single sensor and in a single image. However, this kind of sensor generally generates significant distortions in the images, not allowing a direct application of the methods conventionally used in perspective vision. Our thesis contains a description of two monitoring approaches that take into account these distortions. These methods show the progress of our work during these three years, allowing us to move from person detection to the 3D estimation of its pose. The first step of this work consisted in setting up a person detection algorithm in the omnidirectional images. We proposed to extend the conventional approach for human detection in perspective image, based on the Gradient-Oriented Histogram (HOG) implemented by dallal [Dalal, 2006], in order to adjust it to spherical images. Our approach uses the Riemannian varieties to adapt the gradient calculation for omnidirectional images as well as the spherical gradient for spherical images to generate our omnidirectional image descriptor. The descriptor will be used along with an SVM classifier for decision making. Several experiments have been done using the INRIA image database [Dalal, 2005], as well as our own database. We will introduce the different results obtained with our algorithm for a robust detection of people in omnidirectional images. Subsequently, we set up a 3D tracking system for people with omnidirectional cameras. We have chosen to do a 3D tracking based on a model of the person with 32 degrees of freedom, because we have imposed as a constraint the use of a single catadioptric camera. Our work focused on the implementation of several likelihood functions, based on geodesic distances in the spherical space SO3, as well as on the mapping of the silhouette in the image with the 3D model projected. Our likelihood functions combined with a particlefilter (whose particle propagation model is adapted to spherical space) allows accurate 3D tracking of the person in omnidirectional images. The approach has been validated in real conditions and with different person moves. In this manuscript we have shown that the use of omnidirectional cameras in the field of object detection and tracking can be accurate if we take into account the distortions of this type of sensor.
Complete list of metadata

Cited literature [119 references]  Display  Hide  Download
Contributor : Frédéric Davesne Connect in order to contact the contributor
Submitted on : Monday, April 15, 2019 - 3:24:18 PM
Last modification on : Saturday, May 1, 2021 - 3:50:00 AM


Files produced by the author(s)


  • HAL Id : tel-02100091, version 1


Marouane Boui. Détection et suivi de personnes par vision omnidirectionnelle: approches 2D et 3D. Traitement du signal et de l'image [eess.SP]. Université Paris Saclay; Université d'Evry-Val-d'Essonne; Université Mohammed V, Rabat, 2018. Français. ⟨tel-02100091⟩



Record views


Files downloads