Skip to Main content Skip to Navigation

Synthetic 3D Model-Based Object Class Detection and Pose Estimation

Joerg Liebelt 1
1 LEAR - Learning and recognition in vision
Inria Grenoble - Rhône-Alpes, LJK - Laboratoire Jean Kuntzmann, Grenoble INP - Institut polytechnique de Grenoble - Grenoble Institute of Technology
Abstract : This dissertation aims at extending object class detection and pose estimation tasks on single 2D images by a 3D model-based approach. The work describes learning, detection and estimation steps adapted to the use of synthetically rendered data with known 3D geometry. Most existing approaches recognize object classes for a particular viewpoint or combine classifiers for a few discrete views. By using existing CAD models and rendering techniques from the domain of computer graphics which are parameterized to reproduce some variations commonly found in real images, we propose instead to build 3D representations of object classes which allow to handle viewpoint changes and intra-class variability. These 3D representations are derived in two different ways : either as an unsupervised filtering process of pose and class discriminant local features on purely synthetic training data, or as a part model which discriminatively learns the object class appearance from an annotated database of real images and builds a generative representation of 3D geometry from a database of synthetic CAD models. During detection, we introduce a 3D voting scheme which reinforces geometric coherence by means of a robust pose estimation, and we propose an alternative probabilistic pose estimation method which evaluates the likelihood of groups of 2D part detections with respect to a full 3D geometry. Both detection methods yield approximate 3D bounding boxes in addition to 2D localizations ; these initializations are subsequently improved by a registration scheme aligning arbitrary 3D models to optical and Synthetic Aperture Radar (SAR) images in order to disambiguate and prune 2D detections and to handle occlusions. The work is evaluated on several standard benchmark datasets and it is shown to achieve state-of-the-art performance for 2D detection in addition to providing 3D pose estimations from single images.
Document type :
Complete list of metadata

Cited literature [135 references]  Display  Hide  Download
Contributor : Joerg Liebelt <>
Submitted on : Friday, January 7, 2011 - 11:13:00 AM
Last modification on : Tuesday, February 9, 2021 - 3:16:02 PM
Long-term archiving on: : Friday, April 8, 2011 - 2:58:40 AM


  • HAL Id : tel-00553343, version 1



Joerg Liebelt. Synthetic 3D Model-Based Object Class Detection and Pose Estimation. Human-Computer Interaction [cs.HC]. Institut National Polytechnique de Grenoble - INPG, 2010. English. ⟨tel-00553343⟩



Record views


Files downloads