Space and Time Bounds on Indexing 3D Models from 2D Images
IEEE Transactions on Pattern Analysis and Machine Intelligence - Special issue on interpretation of 3-D scenes—part I
Model-based invariants for 3-D vision
International Journal of Computer Vision
Towards model-based recognition of human movements in image sequences
CVGIP: Image Understanding
Learning flexible models from image sequences
ECCV '94 Proceedings of the third European conference on Computer vision (vol. 1)
Invariants of 6 points from 3 uncalibrated images
ECCV '94 Proceedings of the third European conference on Computer Vision (Vol. II)
Dual Computation of Projective Shape and Camera Positions from Multiple Images
International Journal of Computer Vision
The visual analysis of human movement: a survey
Computer Vision and Image Understanding
Human motion analysis: a review
Computer Vision and Image Understanding
Learning and Recognizing Human Dynamics in Video Sequences
CVPR '97 Proceedings of the 1997 Conference on Computer Vision and Pattern Recognition (CVPR '97)
Wide Baseline Point Matching Using Affine Invariants Computed from Intensity Profiles
ECCV '00 Proceedings of the 6th European Conference on Computer Vision-Part I
| Hi-index | 0.00 |
We present a method for recognition of walking people in monocular image sequences based on extraction of coordinates of specific point locations on the body. The method works by comparison of sequences of recorded coordinates with a library of sequences from different individuals. The comparison is based on the evaluation of view invariant and calibration independent view consistency constraints. These constraints are functions of corresponding image coordinates in two views and are satisfied whenever the two views are projected from the same 3D object. By evaluating the view consistency constraints for each pair of frames in a sequence of a walking person and a stored sequence we get a matrix of consistency values that ideally are zero whenever the pair of images depict the same 3D-posture. The method is virtually parameter free and computes a consistency residual between a pair of sequences that can be used as a distance for clustering and classification. Using interactively extracted data we present experimental results that are superior to those of previously published algorithms both in terms of performance and generality.