Identifying anomalous object types during classification
View Patent ↗Techniques are disclosed for identifying anomaly object types during classification of foreground objects extracted from image data. A self-organizing map and adaptive resonance theory (SOM-ART) network is used to discover object type clusters and classify objects depicted in the image data based on pixel-level micro-features that are extracted from the image data. Importantly, the discovery of the object type clusters is unsupervised, i.e., performed independent of any training data that defines particular objects, allowing a behavior-recognition system to forgo a training phase and for object classification to proceed without being constrained by specific object definitions. The SOM-ART network is adaptive and able to learn while discovering the object type clusters and classifying objects and identifying anomaly object types.
1. A computer-implemented method for identifying anomaly object types during classification of image data captured by a video camera, the method comprising:
receiving a micro-feature vector including multiple micro-feature values, each micro-feature value based on at least one pixel-level characteristic of a foreground patch that depicts a foreground object within the image data;
classifying the foreground object as depicting a first object type corresponding to a first object type cluster of the object type clusters based on the micro-feature vector;
computing a probability density function for the object type clusters;
computing a probability density value for the micro-feature vector;
evaluating a rareness measure of the micro-feature vector, wherein the rareness measure estimates a likelihood of observing the micro-feature vector, based on the probability density function and the probability density value; and
identifying the foreground object as an anomaly object type when the rareness measure is below a specified threshold.
2. The computer-implemented method of claim 1 , wherein the foreground object is identified as the anomaly object type when the rareness measure is greater than a threshold value.
3. The computer-implemented method of claim 1 , wherein the foreground object is classified as depicting the first object type when a distance between the micro-feature vector and a center of the first object type cluster is less than a threshold value.
4. The computer-implemented method of claim 1 , further comprising processing the micro-feature vector by a self-organizing map adaptive resonance theory (SOM-ART) network to discover the object type clusters for the image data.
5. The computer-implemented method of claim 1 , further comprising merging the first object type cluster with a second object type cluster of the object type clusters for the image data when the first object type cluster overlaps the second object type cluster by a specified amount.
6. The computer-implemented method of claim 1 , further comprising:
receiving additional micro-feature vectors; and
removing the first object type cluster when none of the additional micro-feature vectors match the first object type cluster.
7. The computer-implemented method of claim 1 , further comprising adding a new object type cluster when the micro-feature vector does not correspond to any of the object type clusters for the image data.
8. The computer-implemented method of claim 1 , further comprising:
receiving additional micro-feature vectors; and
classifying additional foreground objects based on the additional micro-feature vectors before evaluating the rareness measure of the micro-feature vector.
9. A computer-readable storage medium containing a program which, when executed by a processor, performs an operation for identifying anomaly object types during classification of image data captured by a video camera, the operation comprising:
receiving a micro-feature vector including multiple micro-feature values, each micro-feature value based on at least one pixel-level characteristic of a foreground patch that depicts a foreground object within the image data;
classifying the foreground object as depicting a first object type corresponding to a first object type cluster of the object type clusters based on the micro-feature vector;
computing a probability density function for the object type clusters;
computing a probability density value for the micro-feature vector;
evaluating a rareness measure of the micro-feature vector, wherein the rareness measure estimates a likelihood of observing the micro-feature vector, based on the probability density function and the probability density value; and
identifying the foreground object as an anomaly object type when the rareness measure is below a specified threshold.
10. The computer-readable storage medium of claim 9 , wherein the foreground object is identified as the anomaly object type when the rareness measure is greater than a threshold value.
11. The computer-readable storage medium of claim 9 , wherein the foreground object is classified as depicting the first object type when a distance between the micro-feature vector and a center of the first object type cluster is less than a threshold value.
12. The computer-readable storage medium of claim 9 , further comprising processing the micro-feature vector by a self-organizing map adaptive resonance theory (SOM-ART) network to discover the object type clusters for the image data.
13. The computer-readable storage medium of claim 9 , further comprising:
receiving additional micro-feature vectors; and
classifying additional foreground objects based on the additional micro-feature vectors before evaluating the rareness measure of the micro-feature vector.
14. A system, comprising:
a video input source configured to provide image data;
a processor; and
a memory containing a program, which, when executed on the processor is configured to perform an operation that identifies anomaly object types during classification of image data captured by a video camera, the operation comprising:
receiving a micro-feature vector including multiple micro-feature values, each micro-feature value based on at least one pixel-level characteristic of a foreground patch that depicts a foreground object within the image data;
classifying the foreground object as depicting a first object type corresponding to a first object type cluster of the object type clusters based on the micro-feature vector;
computing a probability density function for the object type clusters;
computing a probability density value for the micro-feature vector;
evaluating a rareness measure of the micro-feature vector, wherein the rareness measure estimates a likelihood of observing the micro-feature vector, based on the probability density function and the probability density value, and
identifying the foreground object as an anomaly object type when the rareness measure is below a specified threshold.
15. The system of claim 14 , wherein the foreground object is identified as the anomaly object type when the rareness measure is greater than a threshold value.
16. The system of claim 14 , wherein the foreground object is classified as depicting the first object type when a distance between the micro-feature vector and a center of the first object type cluster is less than a threshold value.
17. The system of claim 14 , further comprising processing the micro-feature vector by a self-organizing map adaptive resonance theory (SOM-ART) network to discover the object type clusters for the image data.
18. The system of claim 14 , further comprising merging the first object type cluster with a second object type cluster of the object type clusters for the image data when the first object type cluster overlaps the second object type cluster by a specified amount.
19. The system of claim 14 , further comprising:
receiving additional micro-feature vectors; and
removing the first object type cluster when none of the additional micro-feature vectors match the first object type cluster.
20. The system of claim 14 , further comprising adding a new object type cluster when the micro-feature vector does not correspond to any of the object type clusters for the image data.