JPH0729015A - Picture recognition device - Google Patents

Abstract

PURPOSE:To provide a picture recognition device which can securely recognize an unspecified object under' the complicated background of outdoor monitor or the like. CONSTITUTION:A detection means 2 which time-sequentially and sequentially detects at least a mobile object between the pictures at every other picture from the plural pictures 1 which are continuously image-picked up, a judgement means 6 judging whether a detected object continuously moves or not according to whether the detected object is overlapped or not by overlapping the pictures between the plural detected pictures 3 detected by the detection means 2 and an operation means removing the object moving at a prescribed position by subtracting the detected picture before then from the detected picture at prescribed time and detecting the object which continuously moves in a prescribed direction are provided. A moving part at the background without directivity is removed and the object moving in the prescribed direction is recognized.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognition apparatus for recognizing an object under a complicated background such as monitoring.

[0002]

2. Description of the Related Art Conventionally, as a moving object recognition method, there is a pattern matching method between a previously captured reference image and an image at the time of recognition. As a typical example of this method, there is a difference method for obtaining a difference between grayscale values on the same address of both images. This can be easily recognized when the target is moving while the background is fixed, as shown in FIG.

However, the difference method has a problem that it also detects a moving portion of the background such as a sway of a tree existing in the case of an outdoor background. For this reason, various methods have recently been tried in order to eliminate the influence of the shape change of the background. There is a method of detecting and tracking a motion target by combining tracking processing of a change area obtained by the difference method and examining the movement. This method requires a tracking module for each change area, and therefore requires complicated processing such as what standard to assign when there are a plurality of change areas. As another method, there is a method of presetting the size and shape of the target and finding a suitable one in the change detection area. However, since it is necessary to set in advance, it is impossible to detect unspecified objects such as monitoring.

[0004]

As described above, the conventional method has a problem that it is not possible to sufficiently recognize an unspecified object under a complicated background such as monitoring. Therefore, an object of the present invention is to provide a plurality of consecutively imaged objects under the assumption that an object such as a vehicle continuously moves in a certain direction and a moving object in the background undergoes a non-directional change such as a shaking of a tree. Image recognition that detects a moving object detection image from the image, removes the changed part of the background generated on the moving object detection image by a simple method, and can detect the object under a complicated background by detecting the object To provide a device.

[0005]

In order to achieve the above object, the present invention provides at least every second image from a plurality of images continuously picked up by an image recognition apparatus for recognizing an object continuously moving in a fixed direction. There is a detection unit that sequentially detects moving objects in time series, and a determination unit that determines whether or not the detected objects are continuously moving by superimposing images between a plurality of detected images and whether or not the detected objects overlap. An object moving at a fixed position is removed by subtracting a detection image before that from the detection image at the time, and a calculation means for detecting an object continuously moving in a fixed direction is provided. An image recognition device characterized by recognizing an object that is removed and moves in a certain direction.

[0006]

In the image recognition apparatus of the present invention configured as described above, a moving object between images and an image in which the background is changed are detected, and the detected images before that are subtracted from the detected images at a certain time to obtain a fixed position. The moving object is removed, and the image in which the object moving continuously in a fixed direction is detected is output. or,
It is determined whether or not the detected objects are continuously moving from the sequentially created detected images based on whether or not the detected objects overlap, and the target object is recognized by outputting an image in which only the continuously moving objects are detected.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. As shown in FIG. 1, the image recognition apparatus of this embodiment is
Moving object detection means 2 for creating a moving object detection image 3 in which a moving object is detected between at least every other image from a plurality of continuously captured images 1, and a detection image at a certain time point in the detection images sequentially created in time series. From the directionality determining means 4 for creating an image by subtracting the detection images before that and the continuity determining means 6 for determining whether or not the detection objects overlap by superimposing the images between the detection images sequentially created by the moving object detection means. Become.

In the present embodiment thus constructed, as shown in FIG. 2, the moving object detecting means 2 detects the moving object between the images and the moving object detection image 3 in which the background is changed. The directionality determination means 4 subtracts the detected images before that from the detected image at a certain time to remove the object moving at a fixed position, and outputs the detected image 5 to the object moving continuously in a fixed direction. The continuity determining means 6 determines whether or not the detected objects continuously move from the sequentially created detected images, and outputs the image 7 in which only the continuously moving objects are detected. Recognize the object.

Further, as shown in FIG. 3, the moving object detecting means 2 comprises a difference portion 8 and a binarizing portion 9, and the directionality determining means 4
Is composed of a subtracting section 10, and the continuity judging means 6 is composed of an overlapping area detecting section 11.

Here, the difference section 8 outputs the difference between the two images. The binarization unit 9 converts the density of the changed portion into 1 and the density of the unchanged portion into 0. Subtraction unit 10
Is an image obtained by subtracting another difference image from a certain difference image. When the two images are A and B, respectively, the overlapping area detection unit 11 obtains the logical product of the two images and the logical sum of A and the detected object of A that has a portion overlapping with B. The image is output by extracting only the image.

The object recognition is performed with the above configuration. First of all, what is present on the image is as follows.・ Object that moves continuously in a certain direction ・ Still background ・ Noise (noise that occurs in infrared cameras, etc.) ・ Background that changes in a certain area (eg, shaking of trees) (1) Five consecutively captured images (1a to 1e) ) Is input to the difference unit 8 in a pair for each image interval such as 1a and 1c, 1b and 1d, 1c and 1e, and the difference is obtained. In the difference image, the number of pixels having a density equal to or higher than that preset in the binarization unit 9 is 1,
The following pixels are set to 0 to form a binary image, and a moving object detection image (3a to 3c) in which a moving region is detected is sequentially output in time series by extracting a portion where the density has changed by a set value or more. This removes the stationary background.

(2) The fixed-direction moving object detection image 5 obtained by subtracting 3c to 3a out of the three difference binary images sequentially generated in time series by the subtraction unit 10 is output. At this time, since the object continuously moves in a fixed direction, the moving object in 3c always has a portion that does not exist in the previously existing area (area 3a), and therefore there is a portion that is not removed and remains. However, the background that changes in a certain area, such as the shaking of trees, is removed because it is in the area where it existed before.

These fixed-direction moving object detection images are shown in FIG.
As shown in, the following differences can be seen between the target and noise. Since the target moves continuously, there is always a portion where moving objects overlap between images. However, in the case of noise, since there is no continuous movement, there is no overlap between images.

Since it is possible to make a target determination based on whether or not there is an overlapping area from every other image difference image,
In (3), noise removal using this is performed. (3) Next, the moving object detection images 3a and 3b are overlapped by the overlapping area detection unit.
The logical sum of the two logical product images and the moving object detection image 3b is obtained from 11 and only the detected object having a portion overlapping the detected object of the moving object detection image 3a among the detected objects of the moving object detection image 3b is detected. An overlapping area detection image 11a is obtained. Further, the detected overlapping area detection image 11a and the constant-direction moving target detection image 5 are processed by the overlapping area detection unit 11 in the same manner as above,
An overlapping area detection image 11b (corresponding to the target detection image 7) is obtained. As a result, the object is not removed because the moving object always overlaps between the images. However, in the case of noise, there is no overlapping portion, so it is removed.

As a result, as shown in FIG. 4, all but the object on the captured image are removed. As described above, according to the present embodiment, by detecting only an object that moves in a certain direction, a moving portion having no directivity such as a sway of a tree is removed, so that the background is not affected. Even if noise is present, it is removed because it has no continuity. Under the condition that an object such as a vehicle moves in a fixed direction, the object recognition under a complicated background can be performed by the above operation. In this embodiment, every other image is used,
When the amount of movement is small, the present invention can be realized by performing image recognition by taking images at intervals exceeding one image.

[0016]

As described above, according to the present invention, a plurality of consecutively imaged images are acquired on the assumption that the background continuously moves in a certain direction due to fluctuations without directivity such as shaking of trees. It is possible to detect the object from the image without being affected by the fluctuation of the background. Therefore, it is possible to provide an image recognition device having a simple configuration and capable of surely recognizing an unspecified object under a complicated background such as outdoor monitoring.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram showing an embodiment of the present invention.

FIG. 2 is a diagram for explaining the operation of one embodiment of the present invention.

FIG. 3 is a detailed configuration diagram showing an embodiment of the present invention.

FIG. 4 is a view for explaining the detailed operation of the embodiment of the present invention.

FIG. 5 is a diagram showing moving object recognition by a conventional image recognition technique.

[Explanation of symbols]

2 ... Moving object detecting means, 5 ... Direction determining means, 6 ... Continuity determining means.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Office reference number FI technical display location G08B 25/00 510 M 9377-5G

Claims (1)

[Claims] 1. An image recognition device for recognizing an object that continuously moves in a fixed direction, and a detecting means for sequentially detecting a moving object between at least every other image in a time series from a plurality of continuously captured images. , A judgment means for judging whether the detection object is continuously moving by superposing the images among the plurality of detection images detected by the detection means, and the detection image at a certain time An object that moves at a fixed position is removed by subtracting the detected image, and an arithmetic unit that detects an object that continuously moves in a fixed direction is provided. The moving part of the background having no directivity is removed and the object is moved in a fixed direction. An image recognition device characterized by recognizing an object.