Local change detection in video

Patent metadata

Field	Value
Publication number	US-2016277645-A1
Application number	US-201514661406-A
Country	US
Kind code	A1
Filing date	Mar 18, 2015
Priority date	Mar 18, 2015
Publication date	Sep 22, 2016
Grant date	—

Abstract

Techniques related to detecting local change in video are discussed. Such techniques may include determining inlier and outlier keypoints for a current frame of a video sequence based on inlier keypoints from previous frames, detecting a region of local change based on outlier keypoints of the current and previous frames, and providing an indicator of local change based on the detected region of local change.

First claim

What is claimed is:

1 . A method for providing local change detection in video comprising:
determining one or more inlier keypoints and one or more outlier keypoints in a current frame of a video sequence based on one or more second inlier keypoints in one or more previous frames of the video sequence; detecting a region of local change in the video sequence based on the outlier keypoints in the current frame and one or more second outlier keypoints in the one or more previous frames of the video sequence; and providing an indicator of local change based on the detected region of local change in the video sequence.

2 . The method of claim 1 , wherein determining the one or more inlier keypoints and the one or more outlier keypoints comprises, for keypoints of the current frame, comparing matching scores for the keypoints to a matching threshold.

3 . The method of claim 1 , wherein determining the one or more inlier keypoints comprises, for a first inlier keypoint of the one or more inlier keypoints, detecting a matching keypoint from the one or more second inlier keypoints.

4 . The method of claim 3 , wherein the matching keypoint is within a spatial neighborhood of the first inlier keypoint and has a smallest descriptor distance with respect to the first inlier keypoint of any of the one or more second inlier keypoints within the spatial neighborhood.

5 . The method of claim 1 , wherein detecting the region of local change comprises warping the one or more second outlier keypoints to a coordinate system of the current frame and applying a clustering technique to the outlier keypoints and the one or more second outlier keypoints, wherein the region of local change is associated with a cluster of at least some of the outlier keypoints and the one or more second outlier keypoints.

6 . The method of claim 5 , wherein the one or more previous frames comprise multiple previous frames, and wherein an individual outlier keypoint of the one or more second outlier keypoints is associated with a weighting value based on a previous frame of the multiple previous frames associated with the individual outlier keypoint.

7 . The method of claim 1 , wherein the video sequence comprises low resolution video frames at a frame rate of not more than five frames per second.

8 . The method of claim 1 , wherein the indicator of local change comprises at least one of a local motion detected indicator or bounding box location and size data.

9 . The method of claim 1 , further comprising:
estimating global motion associated with the current frame based on correspondences between the one or more inlier keypoints and the one or more second inlier keypoints.

10 . The method of claim 1 , wherein determining the one or more inlier keypoints and the one or more outlier keypoints in the current frame comprises:
determining, for the current frame, a plurality of keypoints and associated local image descriptors; warping the one or more second inlier keypoints to coordinates associated with the current frame; determining, for a first keypoint of the plurality of keypoints, a matching score based on one or more matches between the first keypoint and one or more matching keypoints of the plurality of second inlier keypoints, wherein the one or more matching keypoints are from different frames of the one or more previous frames; and comparing the matching score to a threshold to categorize the first keypoint as an inlier keypoint or an outlier keypoint.

11 . The method of claim 10 , wherein the first keypoint is categorized as an outlier keypoint, the method further comprising:
determining, for the first keypoint, an outlier matching score based on one or more outlier matches between the first keypoint and one or more outlier matching keypoints of the plurality of second outlier keypoints; and categorizing the first keypoint and the one or more outlier matching keypoints as inlier keypoints based on a comparison of the outlier matching score to a second threshold.

12 . The method of 1 , further comprising:
performing a wake up from a low power state based on the indicator of local change.

13 . A system for providing local change detection in video comprising:
a memory configured to receive a video sequence; and a graphics processing unit coupled to the memory, the graphics processing unit to determine one or more inlier keypoints and one or more outlier keypoints in a current frame of the video sequence based on one or more second inlier keypoints in one or more previous frames of the video sequence, detect a region of local change in the video sequence based on the outlier keypoints in the current frame and one or more second outlier keypoints in the one or more previous frames of the video sequence, and provide an indicator of local change based on the detected region of local change in the video sequence.

14 . The system of claim 13 , wherein the graphics processing unit to determine the one or more inlier keypoints and the one or more outlier keypoints comprises, for keypoints of the current frame, the graphics processing unit to compare matching scores for the keypoints to a matching threshold.

15 . The system of claim 13 , wherein the graphics processing unit to determine the one or more inlier keypoints comprises, for a first inlier keypoint of the one or more inlier keypoints, the graphics processing unit to detect a matching keypoint from the one or more second inlier keypoints.

16 . The system of claim 13 , wherein the graphics processing unit to detect the region of local change comprises the graphics processing unit to warp the one or more second outlier keypoints to a coordinate system of the current frame and apply a clustering technique to the outlier keypoints and the one or more second outlier keypoints, wherein the region of local change is associated with a cluster of at least some of the outlier keypoints and the one or more second outlier keypoints.

17 . The system of claim 13 , wherein the indicator of local change comprises at least one of a local motion detected indicator or bounding box location and size data.

18 . The system of claim 13 , the graphics processing unit further to estimate global motion associated with the current frame based on correspondences between the one or more inlier keypoints and the one or more second inlier keypoints.

19 . The system of claim 13 , further comprising:
an image sensor to attain raw image data associated with a scene; and an image processing unit couple do the image sensor and to provide a frame of the video sequence to the memory based on the raw image data.

20 . At least one machine readable medium comprising a plurality of instructions that, in response to being executed on a device, cause the device to provide local change detection by:
determining one or more inlier keypoints and one or more outlier keypoints in a current frame of a video sequence based on one or more second inlier keypoints in one or more previous frames of the video sequence; detecting a region of local change in the video sequence based on the outlier keypoints in the current frame and one or more second outlier keypoints in the one or more previous frames of the video sequence; and providing an indicator of local change based on the detected region of local change in the video sequence.

21 . The machine readable medium of claim 20 , wherein determining the one or more inlier keypoints and the one or more outlier keypoints comprises, for ke

Assignees

• Intel Corp

Inventors

• Bitouk Dmitri
• Zinevich Artem

Classifications

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06F18/2433

  Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection · CPC title

• G06F18/22

  Matching criteria, e.g. proximity measures · CPC title

• G06F18/23

  Clustering techniques · CPC title

• H04N19/139

  Analysis of motion vectors, e.g. their magnitude, direction, variance or reliability · CPC title