Real-time action detection and classification
The present invention relates to a method and system for creating a strong classifier based on motion patterns wherein the strong classifier may be used to determine an action being performed by a body in motion. When creating the strong classifier, action classification is performed by measuring similarities between features within motion patterns. Embodiments of the present invention may utilize candidate part-based action sets and training samples to train one or more weak classifiers that are then used to create a strong classifier.
1. A computerized method for creating a classifier, comprising the steps of:
receiving, by a computer, a training sample set and a candidate part-based action set;
creating, by the computer, one or more weak classifier sets based on the training sample set and the candidate part-based action set wherein the weak classifier sets have assigned error rates based on a comparison of the training sample set and the candidate part-based action set; and
creating, by the computer, a strong classifier based on the one or more weak classifier sets by combining a first weak classifier with a lowest assigned error rate from a selected first weak classifier set with a second weak classifier with a lowest assigned error rate from a selected second weak classifier set.
2. The computerized method of claim 1 , wherein creating the one or more weak classifier sets based on the training sample set and the candidate part-based action set, further comprises:
selecting a first candidate part-based action from the candidate part-based action set;
utilizing a first weak classifier to compare the first candidate part-based action with one or more of the training samples included in the training sample set;
assigning an error rate to the first weak classifier as a result of the comparison of the selected candidate part-based action with the one or more training samples; and
adding the first weak classifier to a first weak classifier set from the one or more weak classifier sets;
selecting a second candidate part-based action from the candidate part-based action set;
utilizing a second weak classifier to compare the second candidate part-based action with one or more of the training samples included in the training sample set;
rating the second weak classifier as a result of the comparison of the selected candidate part-based action with the one or more training samples; and
adding the second weak classifier to the first weak classifier set.
3. The computerized method of claim 1 , wherein the first weak classifier utilizes an exemplar feature of the first candidate part based action and an exemplar feature of the one or more training samples when comparing the first candidate part based action with the one or more training samples.
4. The computerized method of claim 1 , further comprising the step of:
identifying a motion pattern through the use of the strong classifier.
5. A non-transitory computer-readable storage medium configured to store computer code for implementing a method for creating a classifier, wherein the computer code comprises:
code for receiving a training sample set and a candidate part-based action set;
code for creating one or more weak classifier sets based on the training sample set and the candidate part-based action set where the weak classifier sets have assigned error rates based on a comparison of the training sample set and the candidate part-based action set; and
creating, by the computer, a strong classifier based on the one or more weak classifier sets by combining a first weak classifier with a lowest assigned error rate from a selected first weak classifier set with a second weak classifier with a lowest assigned error rate from a selected second weak classifier set.
6. The computer-readable storage medium of claim 5 , wherein the code for creating the one or more weak classifier sets based on a training sample set and the candidate part-based action set, further comprises:
code for selecting a first candidate part-based action from the candidate part-based action set;
code for utilizing a first weak classifier to compare the first candidate part-based action with one or more of the training samples included in the training sample set;
code for assigning an error rate to the first weak classifier as a result of the comparison of the selected candidate part-based action with the one or more training samples; and
code for adding the first weak classifier to a weak classifier set from the one or more weak classifiers sets;
code for selecting a second candidate part-based action from the candidate part-based action set;
code for utilizing an second weak classifier to compare the second candidate part-based action with one or more of the training samples included in the training sample set;
code for rating the second weak classifier as a result of the comparison of the selected candidate part-based action with the one or more training samples; and
code for adding the second weak classifier to the first weak classifier set.
7. The computer-readable storage medium of claim 5 , wherein the first weak classifier utilizes an exemplar feature of the first candidate part based action and an exemplar feature of the one or more training samples when comparing the first candidate part based action with the one or more training samples.
8. The computer-readable storage medium of claim 5 , further comprising:
code for identifying a motion pattern through the use of the strong classifier.
9. A system for creating a classifier, comprising:
a computer configured to:
receive a training sample set, from a first database and a candidate part-based action set, from a second database;
create one or more weak classifier sets based on the training sample set and the candidate part-based action set where the weak classifier sets have assigned error rates based on a comparison of the training sample set and the candidate part-based action set; and
create a strong classifier based on the one or more weak classifier sets by combining a first weak classifier with a lowest assigned error rate from a selected first weak classifier set with a second weak classifier with a lowest assigned error rate from a selected second weak classifier set.
10. The system of claim 9 , wherein the computer is further configured to:
select a first candidate part-based action from the candidate part-based action set;
utilize a first weak classifier to compare the first candidate part-based action with one or more of the training samples included in the training sample set;
assign an error rate to the first weak classifier as a result of the comparison of the selected candidate part-based action with the one or more training samples; and
add the first weak classifier to a first weak classifier set from the one or more weak classifiers sets;
select a second candidate part-based action from the candidate part-based action set;
utilize an second weak classifier to compare the second candidate part-based action with one or more of the training samples included in the training sample set;
rate the second weak classifier as a result of the comparison of the selected candidate part-based action with the one or more training samples; and
add the second weak classifier to the weak classifier set.
11. The system of claim 9 , wherein the first weak classifier utilizes an exemplar feature of the first candidate part based action and an exemplar feature of the one or more training samples when comparing the first candidate part based action with the one or more training samples.
12. The system of claim 9 , wherein the computer is further configured to identify a motion pattern through the use of the strong classifier.