Automatic detection of objects in video images

Patent metadata

Field	Value
Publication number	US-10083378-B2
Application number	US-201615191033-A
Country	US
Kind code	B2
Filing date	Jun 23, 2016
Priority date	Dec 28, 2015
Publication date	Sep 25, 2018
Grant date	Sep 25, 2018

Abstract

A machine learning model is configured to detect objects from video images. A system monitors video images to identify particular objects. A deep learning process is utilized to learn a baseline pattern. A change due to movement within a field of view is autonomously detected using the deep learning processing. An action is performed based on the detected change.

First claim

What is claimed is:

1. A method of detecting objects from video images based on deep learning processing, the method comprising:
learning a baseline pattern from baseline scores obtained using the deep learning processing, the baseline pattern comprising feature vectors; and
autonomously detecting a change due to movement within a field of view using the deep learning processing based on a difference between the baseline scores and classification scores of a plurality of subsequent video frames, the detected change increasing or lowering the classification score.

2. The method of claim 1 , in which learning the baseline pattern comprises:
extracting a first set of feature vectors from a first video frame;
extracting a second set of feature vectors from a second video frame;
representing the first and second set of extracted feature vectors as first and second baseline scores, respectively; and
establishing the first and second video frames as the baseline pattern when the first and second baseline scores are similar to one another.

3. The method of claim 2 , further comprising calculating a final baseline score based on an average of the first and second baseline scores.

4. The method of claim 3 , in which detecting the change comprises:
extracting feature vectors from the plurality of subsequent video frames;
calculating a classification score for each subsequent video frame of the plurality of subsequent video frames; and
detecting the change when the classification score for a set number of the plurality of subsequent video frames remains different from the final baseline score.

5. The method of claim 1 , further comprising storing an image for review by a user based on the detected change.

6. The method of claim 1 , further comprising identifying an object detected within the field of view based on the detected change.

7. The method of claim 1 , further comprising performing a user-configured action based on the detected change.

8. The method of claim 1 , further comprising performing an automatic calibration procedure.

9. The method of claim 1 , further comprising training the deep learning processing to identify the change within the field of view by iteratively repeating training.

10. An apparatus for detecting objects from video images based on deep learning processing, comprising:
a memory; and
at least one processor coupled to the memory, the at least one processor configured:
to learn a baseline pattern from baseline scores obtained using the deep learning processing, the baseline pattern comprising feature vectors; and
to autonomously detect a change due to movement within a field of view using the deep learning processing based on a difference between the baseline scores and classification scores of a plurality of subsequent video frames, the detected change increasing or lowering the classification score.

11. The apparatus of claim 10 , in which the at least one processor is further configured to learn the baseline pattern by:
extracting a first set of feature vectors from a first video frame;
extracting a second set of feature vectors from a second video frame;
representing the first and second set of extracted feature vectors as first and second baseline scores, respectively; and
establishing the first and second video frames as the baseline pattern when the first and second baseline scores are similar to one another.

12. The apparatus of claim 11 , in which the at least one processor is further configured to calculate a final baseline score based on an average of the first and second baseline scores.

13. The apparatus of claim 12 , in which the at least one processor is configured to detect the change by:
extracting feature vectors from the plurality of subsequent video frames;
calculating a classification score for each subsequent video frame of the plurality of subsequent video frames; and
detecting the change when the classification score for a set number of the plurality of subsequent video frames remains different from the final baseline score.

14. The apparatus of claim 10 , in which the at least one processor is configured to store an image for review by a user based on the detected change.

15. The apparatus of claim 10 , in which the at least one processor is configured to identify an object detected within the field of view based on the detected change.

16. The apparatus of claim 10 , in which the at least one processor is configured to perform a user-configured action based on the detected change.

17. The apparatus of claim 10 , in which the at least one processor is further configured to perform an automatic calibration procedure.

18. The apparatus of claim 10 , in which the at least one processor is further configured to train the deep learning processing to identify the change within the field of view by iteratively repeating training.

19. A non-transitory computer-readable medium having program code recorded thereon for detecting objects from video images based on deep learning processing, the program code being executed by a processor and comprising:
program code to learn a baseline pattern from baseline scores obtained using the deep learning processing, the baseline pattern comprising feature vectors; and
program code to autonomously detect a change due to movement within a field of view using the deep learning processing based on a difference between the baseline scores and classification scores of a plurality of subsequent video frames, the detected change increasing or lowering the classification score.

20. An apparatus for detecting objects from video images based on deep learning processing, the apparatus comprising:
means for learning a baseline pattern from baseline scores obtained using the deep learning processing, the baseline pattern comprising feature vectors;
means for autonomously detecting a change due to movement within a field of view using the deep learning processing based on a difference between the baseline scores and classification scores of a plurality of subsequent video frames, the detected change increasing or lowering the classification score.

Assignees

• Qualcomm Inc

Inventors

• Chakraborty Avijit

Classifications

• G06V20/52 Primary

  Surveillance or monitoring of activities, e.g. for recognising suspicious objects (recognising microscopic objects G06V20/69) · CPC title

• G06F18/24

  Classification techniques · CPC title

• G06K9/66 Primary

  Physics · mapped topic

• G06N3/08

  Learning methods · CPC title

• G06K9/4671

  Physics · mapped topic