Patent US3620113A successfully refreshed from Google Patents.
US7519201B2
Expired / Lapsed
US
⏱️ 0.0 years remaining
Detecting humans via their pose
Assignee: Honda Motor Co Ltd
•
Filing: 2006-10-26
•
Granted: 2009-04-14
•
Expires: 2026-10-26
Abstract
Abstract A method and system efficiently and accurately detects humans in a test image and classifies their pose. In a training stage, a probabilistic model is derived in an unsupervised or semi-supervised manner such that at least some poses are not manually labeled. The model provides two sets of model parameters to describe the statistics of images containing humans and images of background scenes. In a testing stage, the probabilistic model is used to determine if a human is present in the image, and classify the human's pose based on the poses in the training images. A solution is efficiently provided to both human detection and pose classification by using the same probabilistic model to solve the problems.
Inventors
- 👤 Ming-Hsuan Yang
- 👤 Alessandro Bissacco
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-09 13:16
CPC Classifications
G; G06; G06V; G06V40/00; G06V40/10; G06V40/103; G06V10/00; G06V10/40; G06V10/50; G06V10/507
Claims Analysis
Total Claims: 30
Claim 1
Independent Claim
102 words
A method for human detection and a pose classification comprising: using a computer to perform the steps of: receiving a probabilistic model derived from a set of training images in at least one of an unsupervised training stage or a semi-supervised training stage; generating a set of test image descriptors representing a test image; determining a likelihood that the test image contains a human based on parameters of the probabilistic model and the test image descriptors; and classifying a body pose of a detected human in the test image based on the test image descriptors and the parameters of the probabilistic model.
Claim 2
Independent Claim
102 words
A method for human detection and a pose classification comprising: using a computer to perform the steps of: receiving a probabilistic model derived from a set of training images in at least one of an unsupervised training stage or a semi-supervised training stage; generating a set of test image descriptors representing a test image; determining a likelihood that the test image contains a human based on parameters of the probabilistic model and the test image descriptors; and classifying a body pose of a detected human in the test image based on the test image descriptors and the parameters of the probabilistic model.
Claim 3
Dependent Claim
25 words
The method of claim 1 wherein the training stage comprises an unsupervised training stage wherein human poses in the set of training images are unlabeled.
Claim 4
Dependent Claim
43 words
The method of claim 1 wherein the training stage comprises a semi-supervised training stage wherein at least one human pose in the set of training images is manually labeled and at least one human pose in the set of training images is unlabeled.
Claim 5
Independent Claim
142 words
A method for human detection and a pose classification comprising: using a computer to perform the steps of: receiving a training image from a set of training images; dividing the training image into at least one cell, each cell having at least one pixel; generating a descriptor for each of the at least one cell, the descriptor representing the cell based on a histogram; generating a probabilistic model based on the set of training images, the probabilistic model representing statistics of the set of training images characterized in that contributions of each component of the probabilistic model combine additively; detecting if a human is present in a test image based on the probabilistic model derived from the set of training images; and classifying a pose of a detected human in the test image based on the probabilistic model and the test image.
Claim 6
Independent Claim
142 words
A method for human detection and a pose classification comprising: using a computer to perform the steps of: receiving a training image from a set of training images; dividing the training image into at least one cell, each cell having at least one pixel; generating a descriptor for each of the at least one cell, the descriptor representing the cell based on a histogram; generating a probabilistic model based on the set of training images, the probabilistic model representing statistics of the set of training images characterized in that contributions of each component of the probabilistic model combine additively; detecting if a human is present in a test image based on the probabilistic model derived from the set of training images; and classifying a pose of a detected human in the test image based on the probabilistic model and the test image.
Claim 7
Dependent Claim
51 words
The method of claim 4 wherein generating a probabilistic model comprises computing a first set of model parameters based on a set of positive training images having images of humans and computing a second set of model parameters based on a set of negative training images having images of background scenes.
Claim 8
Dependent Claim
14 words
The method of claim 4 wherein the descriptor comprises a histogram of oriented gradients.
Claim 9
Dependent Claim
58 words
The method of claim 4 wherein generating the descriptor comprises: computing a gradient for each pixel in the cell; quantizing each gradient into a bin based on an orientation of the gradient; grouping the at least one cell into at least one block of overlapping cells; and normalizing the at least one block for brightness and contrast changes.
Claim 10
Dependent Claim
21 words
The method of claim 4 wherein generating the probabilistic model comprises applying Latent Dirichlet Allocation to the set of training images.
Claim 11
Dependent Claim
29 words
The method of claim 4 wherein generating a probabilistic model comprises learning model parameters based on at least one of Gibbs sampling, Rao-Blackwellised sampling, mean field, or variational inference.
Claim 12
Independent Claim
139 words
A method of human detection and a pose classification comprising: using a computer to perform the steps of: receiving a probabilistic model derived from a set of training images in at least one of an unsupervised training stage or a semi-supervised training stage; receiving a test image; representing the test image by a test image descriptor, wherein the test image descriptor is based on training image descriptors used in the training stage; computing a ratio based on the test image and the probabilistic model, the ratio comprising a probability that the test image comprises a human; comparing the ratio to a predetermined threshold value; outputting a human detection signal if the ratio is greater than the threshold value; and classifying a pose of a detected human in the test image based on the probabilistic model and the test image.
Claim 13
Independent Claim
139 words
A method of human detection and a pose classification comprising: using a computer to perform the steps of: receiving a probabilistic model derived from a set of training images in at least one of an unsupervised training stage or a semi-supervised training stage; receiving a test image; representing the test image by a test image descriptor, wherein the test image descriptor is based on training image descriptors used in the training stage; computing a ratio based on the test image and the probabilistic model, the ratio comprising a probability that the test image comprises a human; comparing the ratio to a predetermined threshold value; outputting a human detection signal if the ratio is greater than the threshold value; and classifying a pose of a detected human in the test image based on the probabilistic model and the test image.
Claim 14
Independent Claim
106 words
A system for human detection and a pose classification comprising: receiving means for receiving a probabilistic model derived from a set of training images in at least one of an unsupervised training stage or a semi-supervised training stage; generating means for generating a set of test image descriptors representing a test image; determining means for determining a likelihood that the test image contains a human based on parameters of the probabilistic model and the test image descriptors; and classifying means for classifying a body pose of a detected human in the test image based on the test image descriptors and the parameters of the probabilistic model.
Claim 15
Independent Claim
106 words
A system for human detection and a pose classification comprising: receiving means for receiving a probabilistic model derived from a set of training images in at least one of an unsupervised training stage or a semi-supervised training stage; generating means for generating a set of test image descriptors representing a test image; determining means for determining a likelihood that the test image contains a human based on parameters of the probabilistic model and the test image descriptors; and classifying means for classifying a body pose of a detected human in the test image based on the test image descriptors and the parameters of the probabilistic model.
Claim 16
Dependent Claim
25 words
The system of claim 11 wherein the training stage comprises an unsupervised training stage wherein human poses in the set of training images are unlabeled.
Claim 17
Dependent Claim
43 words
The system of claim 11 wherein the training stage comprises a semi-supervised training stage wherein at least one human pose in the set of training images is manually labeled and at least one human pose in the set of training images is unlabel.
Claim 18
Independent Claim
157 words
A system for human detection and a pose classification comprising: training image receiving means for receiving a training image from a set of training images; image dividing means for dividing the training image into at least one cell, each cell having at least one pixel; descriptor generating means for generating a descriptor for each of the at least one cell, the descriptor representing the cell based on a histogram; model generating means for generating a probabilistic model based on the set of training images, the probabilistic model representing statistics of the set of training images characterized in that contributions of each component of the probabilistic model combine additively; detecting means for detecting if a human is present in a test image based on the probabilistic model derived from the set of training images; and classifying means for classifying a pose of a detected human in the test image based on the probabilistic model and the test image.
Claim 19
Independent Claim
157 words
A system for human detection and a pose classification comprising: training image receiving means for receiving a training image from a set of training images; image dividing means for dividing the training image into at least one cell, each cell having at least one pixel; descriptor generating means for generating a descriptor for each of the at least one cell, the descriptor representing the cell based on a histogram; model generating means for generating a probabilistic model based on the set of training images, the probabilistic model representing statistics of the set of training images characterized in that contributions of each component of the probabilistic model combine additively; detecting means for detecting if a human is present in a test image based on the probabilistic model derived from the set of training images; and classifying means for classifying a pose of a detected human in the test image based on the probabilistic model and the test image.
Claim 20
Dependent Claim
66 words
The system of claim 14 wherein the model generating means for generating a probabilistic model comprises first parameters computing means for computing a first set of model parameters based on a set of positive training images having images of humans and second parameters computing means for computing a second set of model parameters based on a set of negative training images having images of background scenes.
Claim 21
Dependent Claim
14 words
The system of claim 14 wherein the descriptor comprises a histogram of oriented gradients.
Claim 22
Dependent Claim
77 words
The system of claim 14 wherein the descriptor generating means for generating the descriptor comprises: gradient computing means for computing a gradient for each pixel in the cell; quantization means for quantizing each gradient into a bin based on an orientation of the gradient; cell grouping means for grouping the at least one cell into at least one block of overlapping cells; and normalization means for normalizing the at least one block for brightness and contrast changes.
Claim 23
Dependent Claim
31 words
The system of claim 14 wherein the model generating means for generating the probabilistic model comprises Latent Dirichlet Allocation means for applying Latent Dirichlet Allocation to the set of training images.
Claim 24
Dependent Claim
38 words
The system of claim 14 wherein the model generating means for generating a probabilistic model comprises parameter learning means for learning model parameters based on at least one of Gibbs sampling, Rao-Blackwellised sampling, mean field, or variational inference.
Claim 25
Independent Claim
160 words
A system for human detection and a pose classification comprising: model receiving means for receiving a probabilistic model derived from a set of training images in at least one of an unsupervised training stage or a semi-supervised training stage; test image receiving means for receiving a test image; test image descriptor means for representing the test image by a test image descriptor, wherein the test image descriptor is based on training image descriptors used in the training stage; ratio computing means for computing a ratio based on the test image and the probabilistic model, the ratio comprising a probability that the test image comprises a human; ratio comparing means for comparing the ratio to a predetermined threshold value; signal output means for outputting a human detection signal if the ratio is greater than the threshold value; and classifying means for classifying a pose of a detected human in the test image based on the probabilistic model and the test image.
Claim 26
Independent Claim
160 words
A system for human detection and a pose classification comprising: model receiving means for receiving a probabilistic model derived from a set of training images in at least one of an unsupervised training stage or a semi-supervised training stage; test image receiving means for receiving a test image; test image descriptor means for representing the test image by a test image descriptor, wherein the test image descriptor is based on training image descriptors used in the training stage; ratio computing means for computing a ratio based on the test image and the probabilistic model, the ratio comprising a probability that the test image comprises a human; ratio comparing means for comparing the ratio to a predetermined threshold value; signal output means for outputting a human detection signal if the ratio is greater than the threshold value; and classifying means for classifying a pose of a detected human in the test image based on the probabilistic model and the test image.
Claim 27
Dependent Claim
111 words
A computer program product, comprising a computer readable medium storing computer executable code for human detection and a pose classification, the computer executable code performing steps of: receiving a probabilistic model derived from a set of training images in at least one of an unsupervised training stage or a semi-supervised training stage; generating a set of test image descriptors representing the test image; determining a likelihood that the test image contains a human based on parameters of the probabilistic model and the test image descriptors; and classifying a body pose of a detected human in the test image based on the test image descriptors and the parameters of the probabilistic model.
Claim 28
Dependent Claim
111 words
A computer program product, comprising a computer readable medium storing computer executable code for human detection and a pose classification, the computer executable code performing steps of: receiving a probabilistic model derived from a set of training images in at least one of an unsupervised training stage or a semi-supervised training stage; generating a set of test image descriptors representing the test image; determining a likelihood that the test image contains a human based on parameters of the probabilistic model and the test image descriptors; and classifying a body pose of a detected human in the test image based on the test image descriptors and the parameters of the probabilistic model.
Claim 29
Dependent Claim
27 words
The computer program product of claim 21 wherein the training stage comprises an unsupervised training stage wherein human poses in the set of training images are unlabeled.
Claim 30
Dependent Claim
45 words
The computer program product of claim 22 wherein the training stage comprises a semi-supervised training stage wherein at least one human pose in the set of training images is manually labeled and at least one human pose in the set of training images is unlabeled.
References Cited (Backward Citations)
US20030058111A1
Cited
US20030103647A1
Cited
US20050129278A1
Cited
US20050147278A1
Cited
US20050188013A1
Cited
US20050210103A1
Cited
US20030208289A1
Cited
US7194114B2
Cited
US20050226471A1
Cited
WO2005096216A1
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2012-04-13 | 2012-10-13 | 2013-04-14 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2016-04-13 | 2016-10-13 | 2017-04-14 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2020-04-13 | 2020-10-13 | 2021-04-14 | $7700.00 | Window Closed / Lapsed |