Vision-based object detector
View Patent ↗A method for detecting an object includes acquiring a first video of a source domain where a vision-based object detector is trained. Change detection is performed on the first video to collect one or more first localized, unlabeled sample images from the source domain. A second video is acquired of a target domain where the vision-based object detector is used. Change detection is performed on the second video to collect one or more second localized, unlabeled sample images from the target domain. A domain adaptation parameter is determined based on, one or more domain shift properties between the one or more first localized, unlabeled sample images from the source domain and the one or more second localized, unlabeled sample images from the target domain. The vision-based object detector is adapted to be used for object detection in the target domain based on the domain adaptation parameter.
1. A method for detecting an object, comprising:
acquiring a first video of a source domain where a vision-based object detector is trained;
performing change detection on the first video to collect one or more first localized, unlabeled sample images from the source domain;
acquiring a second video of a target domain where the vision-based object detector is used;
performing change detection on the second video to collect one or more second localized, unlabeled sample images from the target domain;
determining a domain adaptation parameter based on one or more domain shift properties between the one or more first localized, unlabeled sample images from the source domain and the one or more second localized, unlabeled sample images from the target domain, wherein determining the domain adaptation parameter comprises converting the one or more first localized, unlabeled sample images from the source domain into corresponding histogram oriented gradient features; and
adapting the vision-based object detector to be used for object detection in the target domain based on the domain adaptation parameter, wherein adapting the vision-based object detector comprises using a linear support vector machine as a classifier for the source domain and the target domain, wherein W′=WG and b′=W (μ S −Gμ T ), and wherein:
W′ represents a linear weighting vector for the target domain,
W represents a linear weighting vector for the source domain,
G represents a domain shift matrix,
b′ represents a bias for the target domain,
μ S represents a sample means of a data set of the histogram oriented gradient features of the one or more first localized, unlabeled sample images from the source domain, and
μ T represents a sample means of a data set of the histogram oriented gradient features of the one or more second localized, unlabeled sample images from the target domain.
2. The method of claim 1 , wherein the object comprises a vehicle, and wherein the source domain, the target domain, or both comprise a street, a parking lot, or a parking garage.
3. The method of claim 1 , wherein the object comprises a person, and wherein the source domain, the target domain, or both comprise a classroom.
4. The method of claim 1 , wherein performing change detection on the first video comprises using a background-subtraction technique, using motion detection, or using color change detection.
5. The method of claim 1 , wherein the target domain comprises a different location than the source domain.
6. The method of claim 1 , wherein the domain adaptation parameter comprises a matrix that transforms a feature space in the source domain into a feature space in the target domain or a matrix that transforms a classifier in the source domain into a classifier in the target domain.
7. The method of claim 1 , wherein determining the domain adaptation parameter further comprises:
performing principal component analysis on the histogram oriented gradient features to identify one or more sub-spaces; and
calculating the domain shift matrix using the one or more sub-spaces.
8. The method of claim 1 , further comprising rotating a classification hyper-plane of the linear support vector machine in the source domain to the target domain using the domain shift matrix.
9. A non-transitory computer-readable medium storing instructions that, when executed by at least one processor of a computing system, cause the computing system to perform operations, the operations comprising:
acquiring a first video of a source domain where a vision-based object detector is trained;
performing change detection on the first video to collect one or more first localized, unlabeled sample images from the source domain;
acquiring a second video of a target domain where the vision-based object detector is used;
performing change detection on the second video to collect one or more second localized, unlabeled sample images from the target domain;
determining a domain adaptation parameter based on one or more domain shift properties between the one or more first localized, unlabeled sample images from the source domain and the one or more second localized, unlabeled sample images from the target domain, wherein determining the domain adaptation parameter comprises converting the one or more first localized, unlabeled sample images from the source domain into corresponding histogram oriented gradient features; and
adapting the vision-based object detector to be used for object detection in the target domain based on the domain adaptation parameter, wherein adapting the vision-based object detector comprises using a linear support vector machine as a classifier for the source domain and the target domain, wherein W′=WG and b′=W (μ S −Gμ T ), and wherein:
W′ represents a linear weighting vector for the target domain,
W represents a linear weighting vector for the source domain,
G represents a domain shift matrix,
b′ represents a bias for the target domain,
μ S represents a sample means of a data set of the histogram oriented gradient features of the one or more first localized, unlabeled sample images from the source domain, and
μ T represents a sample means of a data set of the histogram oriented gradient features of the one or more second localized, unlabeled sample images from the target domain.
10. The non-transitory computer-readable medium of claim 9 , wherein determining the domain adaptation parameter further comprises:
performing principal component analysis on the histogram oriented gradient features to identify one or more sub-spaces; and
calculating the domain shift matrix using the one or more sub-spaces.
11. The non-transitory computer-readable medium of claim 9 , wherein the operations further comprise rotating a classification hyper-plane of the linear support vector machine in the source domain to the target domain using the domain shift matrix.
12. A computing system comprising:
one or more processors; and
a memory system comprising one or more non-transitory computer-readable media storing instructions that, when executed by at least one of the one or more processors, cause the computing system to perform operations, the operations comprising:
acquiring a first video of a source domain where a vision-based object detector is trained;
performing change detection on the first video to collect one or more first localized, unlabeled sample images from the source domain;
acquiring a second video of a target domain where the vision-based object detector is used;
performing change detection on the second video to collect one or more second localized, unlabeled sample images from the target domain;
determining a domain adaptation parameter based on one or more domain shift properties between the one or more first localized, unlabeled sample images from the source domain and the one or more second localized, unlabeled sample images from the target domain, wherein determining the domain adaptation parameter comprises converting the one or more first localized, unlabeled sample images from the source domain into corresponding histogram oriented gradient features; and
adapting the vision-based object detector to be used for object detection in the target domain based on the domain adaptation parameter, wherein adapting the vision-based object detector comprises using a linear support vector machine as a classifier for the source domain and the target domain, wherein W′=WG and b′=W (μ S −Gμ T ), and wherein:
W′ represents a linear weighting vector for the target domain,
W represents a linear weighting vector for the source domain,
G represents a domain shift matrix,
b′ represents a bias for the target domain,
μ S represents a sample means of a data set of the histogram oriented gradient features of the one or more first localized, unlabeled sample images from the source domain, and
μ T represents a sample means of a data set of the histogram oriented gradient features of the one or more second localized, unlabeled sample images from the target domain.
13. The computing system of claim 12 , wherein determining the domain adaptation parameter further comprises:
performing principal component analysis on the histogram oriented gradient features to identify one or more sub-spaces; and
calculating the domain shift matrix using the one or more sub-spaces.
14. The computing system of claim 12 , wherein the operations further comprise rotating a classification hyper-plane of the linear support vector machine in the source domain to the target domain using the domain shift matrix.