JPS62249298A - Security monitor system - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention relates to a security monitoring system that automatically detects human intrusion into a specific area. This invention relates to a highly functional system that does not misrecognize movement of shadows, etc.

[Background of the invention]

The conventional security monitoring system 11 is limited to the method of detecting the presence of an object by providing an optical path or similar means to a specific area and blocking the optical path, which naturally limits the detection area. There was a problem with putting it away. On the other hand, there is a method in which TV cameras are placed in various locations and the images are centrally displayed for human monitoring.
There is a problem in that the number of objects that humans can monitor is naturally limited.

[Purpose of the invention]

An object of the present invention is to provide a system 11 that automatically and broadly detects human intrusion into an area to be monitored by mechanically processing images from a TV camera.

[Summary of the invention]

When it comes to identifying intruders based on images, it is impossible to make a correspondence between the interior position of the aircraft and the real space position using only two-dimensional information, and the approximate size does not match the actual size. There is a layer between them that could be mistaken for humans.

To this end, the basic idea of the present invention is to process two-dimensional information to generate a distance image, and use this distance image to determine the presence or absence of an intruder.

[Embodiments of the invention]

An embodiment of the present invention will be described below with reference to FIG. By determining the corresponding points between images captured by two TV cameras, the three-dimensional position of the object in the image can be determined using the principle of triangulation. Extraction of this distance information can be performed by the method disclosed in Japanese Patent Application No. 59-237600. To detect an intruder from the obtained distance image,
Two types of prior knowledge are used. The first knowledge is that it is an animal body, and the second knowledge is that it has a size greater than a certain value. Furthermore, in addition to these, if the subject is a human, the subject will have a certain range of heat levels, so capturing an image with an infrared camera is effective in determining that the subject is a human.

First, let's talk about how to extract animal bodies. FIG. 2 shows an example of the configuration of the moving object extraction function. Since the area to be monitored has been set in advance, images in inappropriate positions are erased from the obtained distance image based on this territory information.

This is to simplify processing by avoiding extraction of moving objects outside the show window when the area to be monitored is a show window surrounded by glass. In FIG. 2, this is indicated as a filter. Mobile object extraction is
This can be done by performing a type of difference calculation on the filtered range image. That is, a distance image for a situation where there is no intruder may be stored in advance as a standard pattern, and a difference between this standard pattern and an actually measured distance image may be calculated. If an intruder is included in the measured image, the distance value of the pixel corresponding to the intruder will always be smaller than the distance value of the same pixel in the standard pattern.
By performing the following difference calculation, it is possible to capture only the image of the intruder.

Step 1 (Difference determination from standard pattern) if x(
i, j) or xo(i, j): undefined then e
(i, j): undefined else e (i, j) = x (i, j) - xo (i,
j) Step 2 (Animal body extraction) i f I e(i, jN<i 01” e(itj)
: undefined then d (it J) =, 0els
e d(it j)=x(i, j) here xo (
i, l J): Standard pattern X (ITJ) of the vixel (i, j): Actual measurement value of the vixel (it J) d(1, +j): Distance to the moving object at the vixel N, j): Distance image Generation error tolerance In the above algorithm, it is assumed that -" (ly J) + xo (i, j) includes undefined values. This is because there may be pixels for which distance values cannot be calculated depending on the location of the light. When using a distance image in this way, there is a possibility that missing values may be included, but with a simple two-dimensional image, the image itself changes completely depending on the location of the light, whereas with a distance image, there is a possibility that missing values may be included. This has the effect that the images themselves are no longer completely different even though they simply contain missing values.

Next, the intruder determination function will be described. Distance image to the moving object obtained as a result of moving object extraction D=(d (i, j
)) performs the following processing. First, connected components in D can be extracted. Next, the approximate size is estimated from the extracted connected components. Based on the estimated size of the animal, it is determined whether it is an intruder or not, and if it is an intruder, it is notified.

In extracting connected components, first, one pixel is extracted from a distance image based on a certain index (the following discussion will proceed assuming that the index takes the shortest distance to the camera).Next, A norm whose elements are the above distance and the measured distance is set, and pixels whose norm is less than or equal to a certain value are recorded as belonging to one connected component. Such registration is performed one after another, and pixels with distance values are divided into groups. Once the grouping of the first selected pixel is completed, the same process is applied to the next ungrouped pixel that is at the shortest distance. By repeating these steps, extraction of connected components of the range image regarding the moving object, ie, grouping, is achieved. In the above discussion, the norm was defined only on the measured distance image, but the result of dividing a two-dimensional image into regions may be regarded as one component in norm calculation.

The object size can be estimated in a first-order manner based on the connected component extraction results. The distance image includes three-dimensional coordinate information in real space. Therefore, if a three-dimensional ellipsoid is set for the group of pixels extracted as connected components, the size of the object can be expressed by the length of this principal axis. The above method can determine whether an animal is a human by comparing the size of this object with a certain set threshold.
#I Although this process is performed only on images that have been photographed openly, the reliability of detection can be improved by utilizing the temporal continuity of the images. This is, for example, based on two images at time t and +△ with respect to the extracted moving body image, and the correspondence between the images is determined by Horn and Schunk's Optical Flow Determination 2 Artificial Intelligence 17.
(198]) 185-203 (R, to, P,!to
rn and I3. G, 11.5chunk; r)
cterm, intng Opt, 1cal Flow
, Artificial Intelligence
17. (1981) 185-203), estimate the correspondence of connected components between time t and t+Δt, and evaluate the reliability of the extracted connected components based on the right-hand form of continuity.

By using the above-mentioned method of extracting a moving object using a range image and estimating the size of the extracted object, it is possible to directly measure the size of the object without depending on the lighting conditions, which improves security. The monitoring system 11 is extremely effective. In the previous mQ'l, there were no specific regulations regarding the type of TV camera, but depending on the usage environment, figures outside the area to be monitored may be extracted as moving objects within the area to be monitored, or There is a possibility that unnecessary calculation processing may be performed, such as fluctuations being extracted as moving objects. To avoid this, a method can be adopted in which an infrared camera is used to capture images in the form of heat distribution, and from this the moving object is extracted and its size is estimated.

〔Effect of the invention〕

According to the present invention, it is possible to detect intruding objects, including their size, in a wide area captured by a TV camera, so it is possible to economically construct a highly reliable system with a wide range of applicability in security monitoring. effective.

[Brief explanation of drawings]

Figure 1 is an overall configuration diagram of the present invention, Figure 2 is a diagram of an animal fJ1. TV camera Iris Z diagram

Claims (1)

[Claims] 1. A system for detecting a person's intrusion into a specific area consisting of a sensor and a sensor information processing device, which has a function of extracting a distance image from an image obtained from a TV camera, and a function of extracting a distance image from an image obtained from a TV camera. A security monitoring system characterized by having a function of extracting a moving object by using a plurality of images, and a function of determining and notifying that a moving object is a human based on the feature amount of the extracted moving object distance images. 2. The security monitoring system according to item 1, wherein an infrared camera is used as the TV camera.