Joint pose estimator and feature learning for object detection

Joint pose estimator and feature learning for object detection,10.1109/ICCV.2009.5459304,Karim Ali,François Fleuret,David Hasler,Pascal Fua

Joint pose estimator and feature learning for object detection   (Citations: 3)
BibTex | RIS | RefWorks Download
A new learning strategy for object detection is presented. The proposed scheme forgoes the need to train a collection of detectors dedicated to homogeneous families of poses, and instead learns a single classifier that has the inherent ability to deform based on the signal of interest. Specifically, we train a detector with a standard Ad- aBoost procedure by using combinations of pose-indexed features and pose estimators instead of the usual image fea- tures. This allows the learning process to select and com- bine various estimates of the pose with features able to im- plicitly compensate for variations in pose. We demonstrate that a detector built in such a manner provides noticeable gains on two hand video sequences and analyze the perfor- mance of our detector as these data sets are synthetically enriched in pose while not increased in size. 1. Preamble Machine-learning object detection techniques rely on searching for the presence of the target over all scales and locations of a scene. In order to handle complex cases where latent variables modulate changes in appearance, for instance due to rotation or variation in illumination, two strategies have emerged: either building a collection of pose-dedicated classifiers or explicitly visiting the addi- tional latent variables in the same manner as one explores location and scale. We propose a new approach which consists of design- ing a family of pose estimators able to compute meaningful values for the additional latent variables directly from the signal. We allow the learning procedure to automatically
Conference: International Conference on Computer Vision - ICCV , pp. 1373-1380, 2009
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.
    • ...If we relax the binary flow constraint and let f take continuous values in [0; 1], this results into a convex linear programming system, which can be solved optimally 2 ...
    • ...Our features, as those in [1], compute the ratio of edges of a particular orientation within a sub-window of the detector’s r r square of interest, with respect to the total number of edges within the same sub-window...

    Karim Allet al. FlowBoost — Appearance learning from sparsely annotated video

    • ...In recent work, Ali et al. [2] used pose-indexed features to perform detection of articulated objects, integrating such features directly into a boosted cascade, and ¨...

    Piotr Dolláret al. Cascaded pose regression

    • ...Despite a few recent attempts on simultaneous object detection and pose estimation [15, 2], to the best of our knowledge, none of the existing methods is designed for joint human detection and estimation of body and head poses simultaneously...

    Shaogang Gonget al. Learning human pose in crowd

Sort by: