Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
1998-12-30
2002-01-01
Johns, Andrew W. (Department: 2721)
Image analysis
Pattern recognition
Feature extraction
C386S349000
Reexamination Certificate
active
06335985
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention relates to an object extraction apparatus and, more particularly, to an object extraction apparatus for detecting the position of a target object from input moving picture and tracking/extracting a moving object.
An algorithm for tracking/extracting an object in moving picture has conventionally been proposed. This is a technique of extracting only a given object from a picture including various objects and a background. This technique is useful for a process and editing of moving picture. For example, a person extracted from moving picture can be synthesized with another background.
As a method used for object extraction, the region dividing technique using region segmentation of the spatio-temporal image sequence (Echigo and Hansaku, “region segmentation of the spatio-temporal image sequence for video mosaic”, THE 1997 IEICE SYSTEM SOCIETY CONFERENCE, D-12-81, p. 273, September, 1997) is known.
In this region dividing method using region segmentation of the spatio-temporal image sequence, moving picture is divided into small regions according to the color texture in one frame of the moving picture, and the regions are integrated in accordance with the relationship between the frames. When a picture in a frame is to be divided, initial division must be performed. This greatly influences the division result. In this region dividing method using region segmentation of the spatio-temporal image sequence, initial division is changed by using this phenomenon in accordance with another frame. As a result, different division results are obtained, and the contradictory divisions are integrated in accordance with the motion between frames.
If, however, this technique is applied to tracking/extracting of an object in moving picture without any change, a motion vector is influenced by an unnecessary motion other than the motion of the moving object as a target. In many cases, therefore, the reliability is not satisfactorily high, and erroneous integration occurs.
A moving object detecting/tracking apparatus using a plurality of moving object detectors is disclosed in Jpn. Pat. Appln. KOKAI Publication No. 8-241414. For example, this conventional moving object detecting/tracking apparatus is used for a monitoring system using a monitor camera. This apparatus detects a moving object from an input moving picture and tracks it. In this moving object detecting/tracking apparatus, the input moving picture is input to a picture segmenting section, an inter-frame difference type moving object detector section, a background difference type moving object detector section, and a moving object tracking section. The picture segmenting section segments the input moving picture into blocks each having a predetermined size. The division result is sent to the inter-frame difference type moving object detector section and the background difference type moving object detector section. The inter-fame difference type moving object detection section detects the moving object in the input picture by using the inter-frame difference in units of difference results. In this case, to detect the moving object without being influenced by the moving speed of the moving object, the frame intervals at which inter-frame differences are obtained are set on the basis of the detection result obtained by the background difference type moving detector section. The background difference type moving detector section detects the moving object by obtaining the difference between the moving object and the background picture created by using the moving picture input so far in units of division results. An integration processor section integrates the detection results obtained by the inter-frame difference type moving object detector section and the background difference type moving detector section to extract the motion information about the moving object. After the object is detected from each frame, the moving object tracking section makes the correction moving objects on the respective frames correspond to each other.
In this arrangement, since a moving object is detected by using not only an inter-frame difference but also a background difference, the detection precision is higher than that in a case wherein only the inter-frame difference is used. However, owing to the mechanism of detecting an object in motion from overall input moving picture by using an inter-frame difference and background difference, the detection result of the inter-frame difference and background difference are influenced by unnecessary motions other than the motion of the target moving object. For this reason, a target moving object cannot be properly extracted/tracked from a picture with a complicated background motion.
Another object extraction technique is also known, in which a background picture is created by using a plurality of frames, and a region where the difference between the pixel values of the background picture and input picture is large is extracted as an object.
An existing technique of extracting an object by using this background picture is disclosed in “MOVING OBJECT DETECTION APPARATUS, BACKGROUND EXTRACTION APPARATUS, AND UNCONTROLLED OBJECT DETECTION APPARATUS”, Jpn. Pat. Appln. KOKAI Publication No. 8-55222.
According to this technique, the moving picture signal of the currently processed frame is input to a frame memory for storing one-frame picture data, a first motion detection section, a second motion detection section, and a switch. A video signal one frame ahead of the current frame is read out from the frame memory and input to the first motion detection section. The background video signals generated up to this time are read out from the frame memory prepared to hold background pictures and is input to the second motion detection section and the switch. Each of the first and second motion detection section extracts an object region by using, for example, the difference value between the two input video signals. Each extraction result is sent to a logical operation circuit. The logical operation circuit calculates the AND of the two input video data, and outputs it as a final object region. The object region is also sent to the switch. The switch selects signals depending on an object region as follows. For a pixel belonging to the object region, the switch selects a background pixel signal. In contrast to this, for a pixel that does not belong to the object region, the switch selects the video signal on the currently processed frame, and the signal is sent as an overwrite signal to the frame memory. As a result, the corresponding pixel value in the frame memory is overwritten.
According to this technique, as disclosed in Jpn. Pat. Appln. KOKAI Publication No. 8-55222, as the processing proceeds, more accurate background pictures can be obtained. At the end, the object is properly extracted. However, since the background picture is mixed in the object in the initial part of the moving picture sequence, the object extraction precision is low. In addition, if the motion of the object is small, the object picture permanently remains in the background picture, and the extraction precision remains low.
As described above, in the conventional object extraction/tracking method, owing to the mechanism of detecting an object in motion from the overall input moving picture, the detection result of the inter-frame difference and background difference are influenced by unnecessary motions other than the motion of the target moving object. For this reason, a target moving object cannot be properly extracted/tracked.
In the object extraction method using background pictures, the extraction precision is poor in the initial part of a moving picture sequence. In addition, if the motion of the object is small, since a background picture remains incomplete, the extraction precision remains low.
BRIEF SUMMARY OF THE INVENTION
It is an object of the present invention to provide an object extraction apparatus for moving picture which can accurately extract/track a target object without being
Ida Takashi
Sambonsugi Yoko
Watanabe Toshiaki
Azarian Seyed
Johns Andrew W.
LandOfFree
Object extraction apparatus does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Object extraction apparatus, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Object extraction apparatus will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2864322