Detection of moving objects in video data by block matching to derive a region motion vector
AssigneeNEC Corporation
InventorTakayuki Saeki
Life of this patent
Estimating the term…
Abstract
In a video analysis device, a frame separator separates from frame images respresentative of a moving object at least first and second key frames based on either a scene change or a great movement of the object. Responsive to color and edge information, a region divider divides each key frame into region data. A vector detector applies block matching to the key frames to detect block motion vectors and detects in response to the region data of the first key frame a plurality of region motion vectors. Responsive to the region data and the region motion vectors, a correspondence establishing unit establishes correspondence between one of the region data of the first key frame and the region data of the second key frame to identify the object.
Claims29
A video analysis method comprising the steps of:
- separating from a successive of video data two key frames representative of a moving object;
- forming a predetermined number of region data in a first key frame of said two key frames;
- detecting in response to said region data a region motion vector in said first key frame relative to a second key frame of said two key frames;
- and
- establishing a correspondence between said region motion vector and said region data to identify said moving object.
A video analysis device comprising:
- separating means for separating from a succession of frame images two key frames representative of a moving object;
- forming means for forming a predetermined number of region data in a first key frame of said two key frames;
- detecting means responsive to said region data for detecting a region motion vector in said first key frame relative to a second key frame of said two key frames;
- and
- establishing means for establishing a correspondence between said region motion vector and said region data to identify said moving object.
Dependencies are inferred from the claim text.