JP3093346B2 - Moving object detection device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving object detecting apparatus for detecting a moving object such as an intruder in a monitoring area.

[0002]

2. Description of the Related Art Recently, various measures have been considered as security measures against crime prevention in buildings and offices. For example, moving an intruder or an intruding object into a monitoring area using an imaging device such as a TV camera is considered. A moving object detection device that automatically detects an object has been put to practical use.

Conventionally, this type of moving object detecting apparatus has been used to obtain a difference between a plurality of images taken continuously or a difference between a background image without an intruder and a current image. Many methods for detecting an object are used.

However, in this method, noise other than an intruding object, for example, fluctuation of illumination intensity in an imaging range, reflection of light such as a water surface, brightness change in a certain area such as tree sway, or imaging. Since the influence of noise at the time of imaging of the apparatus itself is detected at the same time as the intruder, the intruder and the noise must be clearly separated, in other words, one intruder is detected as one region in the image. It was very difficult to do.

Therefore, conventionally, as a method for solving such a problem, a block integration processing method (Japanese Patent Laid-Open No. 2-59977 and Japanese Patent Laid-Open No. 2-59978) has been proposed. In this method, a corresponding change region is obtained by using an attribute value of the region between two successive change regions detected at two consecutive points in time, and a plurality of change regions detected from the estimated movement route are integrated. To extract a moving object.

However, the change area detected by this method also includes a large number of the aforementioned noises, and one moving object is detected by being divided into unspecified unspecified plural areas. It has not yet reached a fundamental solution.

As a method of solving this, there is a method of using a tracking means capable of tracking a moving object. According to this method, since a moving object is tracked, there is no need to make correspondence between regions. However, the tracking means used for this means is often used for human motion analysis and the like. As a method, a part to be tracked in advance such as a marker (target) attached to an object to be measured (human limbs, etc.) is used. Because of the need to determine, it is impossible to detect and track an unspecified number of intruders in the monitoring area one after another.

On the other hand, as a method of tracking an unspecified object in the monitoring area, a change detecting means for detecting a change area which changes due to the movement of a moving object such as an intruder is provided, and the change detecting means detects the change area. There has been proposed a method in which a tracking means is provided for a change area to track a moving object which is considered to be in the change area.

The tracking means at this time includes a method of tracking a change in brightness (pixel luminance) in the tracking area, a method of tracking a specific shape in the tracking area, a method of tracking the tracking area and its tracking area. There is a method of calculating a cross-correlation value with a surrounding area to determine a moving direction. Also, since a certain tracking means can track only a part of the moving object, after the tracking is completed, the result is analyzed and the tracking means is divided into a set for each moving object (the moving direction of the tracking means and the mutual direction). A method of collecting tracking means into a set based on a positional relationship) is also conceivable, and a method of using a plurality of tracking means as a means for simultaneously detecting an unspecified large number of moving objects in a monitoring area.

[0010] According to these methods, it is possible to cope with a case where the tracking of some of the tracking means fails due to external factors such as noise or occlusion by an object. However, when the moving objects move in a complicated manner, it is very difficult to interpret the result after the end of the tracking, and it is difficult to interpret the results. However, if one tries to track objects moving in different directions using different tracking means, there is a drawback that these tracking means cannot be grouped into the same set, making it impossible to detect a moving object. Was.

[0011]

As described above, in the conventional moving object detection apparatus, when a moving object such as an intruder is tracked and detected by using a plurality of tracking means, it is complicated like a human. When a plurality of moving objects exist in the detection area at the same time, there is a problem that it is impossible to detect each intruder at the same time.

The present invention has been made in view of the above circumstances, and even if an indefinite number of intruding objects, such as humans, which move in a detection area exist in an unspecified number, it is necessary to track these and interpret the results at the same time. It is an object of the present invention to provide a moving object detection device capable of detecting the movements of a plurality of intruders.

[0013]

SUMMARY OF THE INVENTION The present invention provides a change detecting means for detecting a change area from a captured image, a tracking means having at least two tracking elements for tracking the change area detected by the change detection means, a change area. A virtual connection strength between the tracking elements that track the object is determined, and a set of tracking elements having the connection strength not less than a predetermined value is extracted as a candidate for a moving object .
And a result interpreting means for interpreting an extraction result by the candidate extracting means and generating a moving path of the moving object candidate.

[0014]

As a result, according to the present invention, a set of tracking elements connected by a virtual connection having an intensity equal to or greater than a predetermined value is extracted as a candidate for a moving object, and the moving path of the moving object is extracted from these extraction results. When there are multiple intruders in the surveillance area and there are complex movements such as intersecting, or when the same moving object such as a human makes separate movements In this case, the moving path of the moving object can be detected simultaneously with the tracking.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A moving object detecting apparatus according to one embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows a schematic configuration of the embodiment.
In this case, the moving object detection device includes a TV camera 1 that captures an image, a change detection unit 2 that detects a change area of the captured image,
A tracking unit 3 for tracking a change area; a host computer 4 for controlling the tracking unit 3 and interpreting the tracking result; and a display device including a CRT for visualizing the image of the TV camera 1 and the tracking area in the tracking unit 4. 5. A display device 6 composed of a CRT for visualizing the interpretation result of the host computer 4 for interpreting the tracking result. Then, the image captured by the TV camera 1 is sent to the A / D converter 7, digitized, and sent to the change detection unit 2 and the tracking unit 3.

The change detecting section 2 comprises a reference image memory 21, a difference circuit 22, a binarizing circuit 23, and a labeling circuit 24. Here, the reference image memory 21 stores the background image captured by the TV camera 1 when there is no intruding object. The difference circuit 22 is a TV camera 1
The difference between the input current image and the image stored in the reference image memory 21 is obtained. The binarization circuit 23 binarizes the difference result of the difference circuit 22 based on an appropriate threshold value as to whether or not it is a changed portion, and extracts a portion that is largely changed with respect to the reference image. The labeling circuit 24 gives a different label to each connected component as a change area in the difference image that is the result of binarization. Further, the area, the center of gravity position, and the circumscribed rectangle coordinate value are obtained for each connected component.
(Hereinafter, “connected component” is used as a synonym for “variable region”.)

When the labeling circuit 24 detects a connected component having a predetermined area or more, the tracking unit 3 is given a command to track the connected component from the tracking device control unit 41 of the host computer 4. ing. In the present embodiment, the tracking unit 3 is composed of a multiprocessor, and the individual processor portions of the multiprocessor are respectively configured in the tracking device 31, and one tracking device 31 tracks one connected component. When the number of tracking devices 31 is small, a single tracking device 31 may track a plurality of connected components.

Each tracking device 31 is connected to the host computer 4
And a control bus 32 through which commands from the host computer 4 are received and the results of tracking are transferred to the host computer 4. Further, the tracking device 31 inputs an arbitrary part of the moving image from the A / D converter 7 through the image bus 33. The tracking device 31 itself sets, as a tracking region, a rectangular region of an appropriate size including the circumscribed rectangle of the connected component to be tracked specified by the tracking device control unit 41 as a tracking region, and sets the tracking region within the tracking region. And tracking at least a portion of a moving object such as an intruder.

Here, as a tracking method of the tracking device 31, for example, a method of tracking a change area using a temporal change of pixel luminance in a tracking area is used. In this case, when no change is detected in the pixel luminance in the tracking area or when the moving amount of the moving object is 0, the tracking is determined to have failed, and the tracking is stopped. On the other hand, the tracking device control unit 41 that controls the tracking device 31 has a table memory (not shown) for recording the status of each tracking device 31, and what state the tracking device 31 is currently in. (Tracking, stopped, etc.). For example, when the tracking device 31 stops tracking halfway, the description of the location indicating the status of the tracking device 31 changes from tracking to stopping.

The host computer 4 has a candidate extracting unit 42, a result analyzing unit 43, and a knowledge providing unit 44 in addition to the tracking device control unit 41. Then, the tracking device control unit 41
Controls the tracking devices 31. In this case, the change detection unit 2 is always operated, and when a change area is detected, the tracking device 31 that has not been tracked is assigned to the position. To start. However, if an intruding object exists in the detection area, a change area caused by the intruding object is detected by the change detection unit 2, so that a part already tracked by another tracking apparatus 31 is newly added to another tracking apparatus. It is necessary to prevent tracking at 31. For this reason, after receiving the position of the center of gravity and the circumscribed rectangle of the change area from the change detection unit 2, the result of the tracking device 31 that is currently performing the tracking process is checked. The tracking process is not started for the change area. In this case, the tracking device 31 performing the tracking process sends the time-series data (tracking result) of its own travel route to the tracking device control unit 41 sequentially or after a predetermined amount is accumulated. The tracking device controller 41 collects the results of all the tracking devices 31 and sequentially or after accumulating a predetermined amount of the results, performs a process of extracting a candidate for a moving object.
2. (Hereinafter, "candidate" is used as a synonym for "moving object candidate.")

The display device 5 visualizes the operation of the tracking unit 3. The display device 5 here outputs the tracking area set in each tracking device 31 by the display controller 8 as, for example, a frame.
The image input from the camera 1 is displayed at the same time.

FIG. 2 shows a display example on the screen of the display device 5. In this figure, the screens at two different points in time are shown on the same screen. The two broken lines of the figure in the figure represent the positions of the same intruder at two time points, and the right broken line represents the state after the tracking processing has been performed from the left side several times. . Also, in the figure, the tracking devices (31-1, 31-2, 31) that track the same intruder
Of the -3), the tracing process has failed halfway and the tracing is stopped (31-3), and the tracing is newly arranged and the tracing is started after the change area is detected (31-).
And (3-5) arranged in a change area derived from another object.

By the way, even in the method considered so far, a method of separating tracking devices that track one moving object as a set as a tracking result determination process is described. It is to imitate what is moving to one set. Even in this method, objects arranged in a change region not derived from a certain degree of motion, such as a tracking device (31-5) arranged in a change region derived from another object, should not be included in the set. Is possible. However, in general, when a non-rigid body such as a human moves, each part of the object mostly moves separately, and a tracking device (31-1, 31-2, 31) that tracks the same person is generally used. In the case where -3) performs different exercises at each time point, the above method cannot cope.

On the other hand, in the present embodiment, by simultaneously referring to the moving direction of the tracking device and the mutual positional relationship, a candidate for a moving object that can be determined to be an intruder is sent at the same time as the result of the tracking process is sent. The candidate is extracted and then always interpreted along the movement path of the candidate, and by verifying the validity of the candidate, it is possible to stably detect even a target object that moves complicatedly. Next, FIG. 3 shows a flowchart of the candidate extraction processing by the host computer 4.

In this case, in the candidate extracting unit 42 of the host computer 4, the tracking information is output from the tracking device 31 (step 301), and when the result is received from the tracking device control unit 41 (step 302), the two tracking devices 31
A virtual connection is assumed between them (step 303). In this case, the tracking device 31 that has not performed the tracking process is excluded.

Next, the strength of the connection is calculated (step 3).
04). The strength of this connection is calculated using an evaluation function that takes a value that increases when the movement paths of the tracking devices 31 are very similar (the direction of motion and the speed are close). This evaluation function is prepared in advance in the candidate extraction unit 42, and examples thereof include the following expression. s = Kp (t) · F (p1 (t), p2 (t)) + Kv (t) · G (v1 (t), v2
(T))

In this equation, s is the bond strength, t is time, p1 (t), p2 (t), v1 (t), and v2 (t) are time t
Represents the position of the tracking device 31-1, the position of the tracking device 31-2, the speed of the tracking device 31-1, and the speed of the tracking device 31-2. F and G are predetermined functions using the position and speed of the tracking device 31 as arguments, respectively, and Kp (t) and Kp (t)
v (t) is a coefficient of the functions F and G that can change with time. When the connection strength is calculated in this way, only the connection having a strength equal to or higher than a predetermined threshold is left, and other connections are deleted (step 305).

Next, the candidate extracting unit 42 performs a process of separating the tracking device groups connected by the remaining connections into one set and separating them into each set (step 306). A candidate list at the present time as a candidate is created (step 307). Here, a plurality of intruders exist in the monitoring area, and each intruder is tracked by the tracking device 3.
1 is tracking, a moving object candidate for each intruder is extracted, and a candidate list is created at each time when the tracking result is obtained, and this candidate list is stored in the host computer 4. It is accumulated and a candidate list time series is created. If the time range of the tracking result to be processed is only the current time, only the current time is included in the time series of the candidate list. After the tracking process is performed, the candidate list is sequentially created and the result is interpreted. This is sent to the unit 43. Furthermore, if the time range to be processed includes a plurality of time points, the candidate list at the plurality of time points is accumulated in the process of creating a time series of the candidate list (step 308), and the result is interpreted as time-series data. It is sent to the unit 43. In this case, a process of transferring the result to the display device 6 is also performed.

When the processing of the time range to be processed is completed, the processing returns to step 302, and the same processing as described above is repeated. (Hereinafter, "candidate list string" is used as a synonym for "time-series data of the candidate list over time.")

Incidentally, the tracking process by the tracking unit 3 is not always successful, and in an extreme case, what is being tracked may be replaced on the way. When such a situation occurs in a certain time range, the time series data of the candidate list obtained from the tracking result in this time range indicates that some of the tracking devices 31 included in one candidate are tracked in the middle. Is failed and is excluded from the candidates, and the newly arranged tracking device 31 is added to the candidates in the middle and recorded. In such a case, the time-series data of the candidate list output from the candidate extracting unit 42 does not directly represent the moving path of the moving object.

In this embodiment, the candidate list sequence is sent to the result interpreting unit 43, and the knowledge list provided by the knowledge providing unit 44 is referred to the candidate list sequence so as to represent the moving path of the moving object. A process of converting the data into a data string is performed. In this case, the result interpreting unit 43 is operated in parallel with the candidate extracting unit 42, and the knowledge providing unit 44 is operated in parallel therewith, so that the result interpreting unit 43 and possibly the candidate extracting unit 42 refer to knowledge. When requested, knowledge can be provided immediately. Next, FIG.
5 shows a flowchart of a result interpretation process performed by the result interpretation unit 43.

In this case, the candidate list string created by the candidate extracting unit 42 is received (step 402), and the first one is set as the current candidate list (step 40).
3). Next, the candidates are associated with each other between the current and next candidate lists (step 404). The candidate list at the present time may include candidates before the previous time point at which the processing is suspended under the conditions described later. As the association method here, for example, a method of associating with the one that has the largest number of tracking devices 31 that are included in common is the simplest method. Further, how the group of tracking devices 31 included in the candidate changes with time (continuity of the tracking device), the tracking device 3
Information such as whether the movement route of the candidate 1 or the candidate is smooth (continuity of motion) and how the mutual positional relationship of the tracking devices 31 in the candidate changes with time (continuity of structure) A method may be used in which candidates extracted from candidates and interpreted as being more continuous are associated with each other.

The knowledge used to associate these candidates is
Instead of being recorded in the result interpreting unit 43, it may be recorded in the knowledge providing unit 44 and referred to sequentially. In this way, the interpretation of the tracking result progresses, and the knowledge about the intruding object obtained during that time can be newly registered in the knowledge providing unit 44. If the new knowledge is necessary, the candidate extracting unit 42 can also be provided.

If there is a correspondence between two consecutive candidates, it is determined as YES in step 405, and the fact that the two candidates are the same is recorded in the candidate list (step 406). . On the other hand, if there is no correspondence between the candidates at the two time points, NO is determined in step 405, and the knowledge about the type of the candidate recorded in the knowledge providing unit 44 is referred to (step 407). Is recorded as the most applicable one (step 408). For example, when it becomes impossible to correspond to a candidate near the boundary of the monitoring area or in an area recorded as an entrance, the candidate can be recorded as having left the monitoring area. In addition, candidates that cannot be dealt with in places other than the above are temporarily recorded as invisible (becoming shadows and intersect with other moving objects) and are temporarily suspended, so It is possible to deal with candidates. Further, those which are registered in the knowledge providing unit 44 but cannot be interpreted are recorded as uninterpretable.

When the processing is completed for all the candidate lists at the time of the current processing, and YES is determined in step 409, if the current time is not the last time of the candidate list row, the next time is reviewed with the current time. If NO is determined in 410, the above processing is repeated.

When the association is completed for all the points in the candidate list string and YES is determined in step 409,
After re-recording a plurality of candidates that are associated with time continuation and are recorded as the same candidate as the same moving object candidate (step 411), if the next processing such as result display exists, Then, processing for transferring the result to the display device 6 is also performed (step 412). When the processing to be performed on all data is completed, the process returns to step 402, accepts a new candidate list string, and repeats the above processing.

The candidate extracting unit 42 and the result interpreting unit 4
FIG. 5 schematically shows the result obtained by the processing in FIG.
Can be expressed as The flow of the processing will be further described with reference to FIG.

In this case, in the figure, the state of the screen of the display device 6 at a total of four points in time when the tracking process is performed when two intruders are present in the detection area (51-a to 54-a).
And the connections assumed between the tracking devices 31 at each time point are schematically shown (51-b to 54-b and 52-c to 54-c).

First, the processing in 51-b to 54-b will be described. In this case, there are a total of three tracking devices 31 for performing the tracking process, and one of these tracking devices 31 (m1 in FIGS. 51-a to 54-a) passes a long distance in the detection area. The intruder is being tracked, and the remaining two tracking devices 31 (m2, m3 in the diagrams 51-a to 54-a)
Is simultaneously tracking intruders passing near the other TV camera.

In the candidate extraction process in the candidate extraction unit 42, as described above, the functions F and G included in the evaluation function correspond to the positions and velocities (direction and size) of the two tracking devices 31, respectively. A function for calculating the coupling strength that returns a large value when it is close.
(t) is a value close to 1 at the beginning of tracking, but rapidly decreases as the tracking process progresses, and changes to approach 0,
Conversely, Kv (t), which is a coefficient of the function G, is set to 0 at the beginning of tracking, but increases as the processing proceeds and approaches 1.

In this way, the coupling strength is determined by the distance between the tracking devices 31 at the beginning of tracking. In 51-b indicating the initial state of the tracking, the position of each tracking device 31 at the time when the tracking device 31 is arranged (the tracking area is a rectangle in 51-b) and the tracking device 31
Among the connections assumed between them, those having an intensity equal to or higher than a predetermined threshold value (s1 in 51-b, hereinafter, the thickness of a line segment indicates the intensity) are shown. That is, in the early stage of such tracking, since the movement information has not been obtained from the tracking device 31, the coupling strength is calculated only by the distance between the tracking devices 31. In this example, only the connection between m2 and m3, which are close to each other, is calculated to have a large intensity exceeding the predetermined threshold, and only s1 is connected.

Therefore, as a candidate extraction result at this point, a total of two candidates for a moving object are output, including the one including only the tracking device m1 and the one including m2 and m3 connected by the connection s1. Each of these candidates corresponds to an intruder, and correct moving object candidates are extracted.

Next, when the tracking processing in the tracking device 31 proceeds as shown at the time (52-b), the tracking device m1 and other m
2. The distance from m3 is shorter. Here, if the coefficient of the evaluation function is the same as that in the initial state of the tracking, the coupling strength between these tracking devices 31 is also calculated to be larger than the threshold value, and each should be coupled. is there. However, at the same time, since the coefficient Kp of the evaluation function has changed to a small value and Kv has changed to a large value, the calculation of the coupling strength is based on whether the direction and magnitude of the motion are closer than the distance between the tracking devices 31 or not. Is effective. Therefore, in this example (52-b), only the same s1 as the previous time point is determined to be a connection having a large value equal to or larger than the threshold value, and a correct moving object candidate corresponding to the intruder is extracted as in the previous time point. Is done.

The same applies to the subsequent time points (53-b, 54-b). Even if the distance between the tracking devices 31 is further reduced, correct moving object candidates are extracted. In this case, since Kv becomes a larger value, if the tracking device 31 continues the same exercise,
The strength of the already connected connection becomes larger as the tracking processing proceeds.

Next, the result interpreting unit 43 receives the result obtained by the candidate extraction processing as a time-sequential list, associates time-continuous candidates from the same tracking device 31, and further associates the associated candidates with the same candidate. The processing is re-interpreted as indicating a moving object, and finally a process is performed to obtain the moving path of the moving object in the detection area.

In the above-described example, a moving object can be correctly detected even if no knowledge is recorded in the result interpretation unit 43 or the knowledge providing unit 44 is not provided. An example in which a moving object can be correctly detected using the method will be described with reference to 52-c to 54-c in FIG.

In this example, the tracking process is started from the above-mentioned time point 52-a. For example, when a distant intruder is detected at a location where an intruder is detected, two intruders are detected at the time of detection. This is the case where the distance of the person is short. In this case, 52- at the time when the tracking device 31 is arranged
In a, since the distances between the tracking devices 31 between m1 and m2 and between m2 and m3 are substantially equal, the connection strength of each tracking device 31 is also calculated to be substantially the same value exceeding the threshold value (52). -C, s2, s3 in the figure). At this time, since the three tracking devices 31 are connected, as a result of the candidate extraction processing, only one set including all three tracking devices 31 is output as a moving object candidate, and the moving object candidate extraction is incorrect. It can be said that it has.

At the next time point (53-c),
Although the position of the tracking device 31 has changed due to the progress of the tracking process, each of the tracking devices 31 connected at the above-mentioned time point has been changed.
The distance between them has not changed much. Further, at this time, the coefficient Kp of F is small, and the coefficient Kv of G is large. Conversely, even if the distance between the tracking devices 31 does not change so much, the coupling strength performs the same motion. Tracking device 3
The intensity of s2, s3: s2 in FIG. 53-c is small, and the intensity of s3 is largely changed. But still all three tracking devices 3
Because 1 is combined, only one candidate is extracted and remains incorrect. At the next time point (54-c) at which the tracking processing further proceeds, Kv further increases, and only the tracking device 31 having a close speed is connected (s3 in FIG. 53-c). Two outputs including only m1 and those including m2 and m3 are output, and correct candidates corresponding to the moving object are output.

The candidate list obtained by the above candidate extraction processing is as follows.
At first, there is only one candidate, but it splits in the middle and at the end there are two candidates. In this state, no moving object is correctly detected. However, if a condition relating to the connection such as “a connection in which the strength decreases to a certain value and falls below the threshold value is an erroneous connection” is recorded as knowledge in the result interpreting unit 43, the connection is determined from this knowledge. s2 can be interpreted as an erroneous combination, and the candidate can be re-divided into correct ones corresponding to each moving object. Note that such knowledge may be recorded in the knowledge providing unit 44, and only the knowledge that has been requested by the result interpreting unit 43 to be referred to may be provided. Note that the present invention is not limited to the above-described embodiment, and can be appropriately modified and implemented without departing from the scope of the invention.

In the above-described embodiment, the change detection portion is realized by a dedicated circuit, the tracking portion is realized by a multiprocessor system, and the tracking device control, candidate extraction, result interpretation, and knowledge provision are realized by the host computer. This enables high-speed processing and can cope with a case where the target object moves fast. However, depending on the application, all or part of the DSP
(Digital Signal Processor), a microprocessor, or another digital computer.

For example, in the above-described embodiment of the change detecting section 2, the change area is extracted by a difference from a previously created reference image. It may be performed by differential (difference) calculation.
In this case, the possibility of erroneous change area extraction due to environmental change is reduced. However, in this case, since the contour portion of the object is detected as a change area, it is necessary to appropriately divide the detected change area and set the location of the tracking means. Even when the reference image is used, the reference image may be recreated at appropriate intervals in order to prevent erroneous detection of a change due to environmental fluctuation. Further, in the above embodiment, the threshold value for binarizing the difference image is set to a fixed value without changing, but if this threshold value is made variable according to the density of each part of the image, A reliable change can be detected.

In the candidate extraction processing performed by the candidate extraction unit 42, the evaluation function used in the candidate extraction processing includes the movement of the tracking device such as the position and speed in the image of each tracking device. Although only information related to the route is used, feature amounts such as colors and shapes obtained from the tracking area of the tracking device may be added. For example, consider the case where a plurality of tracking devices are simultaneously tracking the body of an intruder wearing red clothing, and at this time, after extracting color information from the tracking area of each tracking device, this is the same. Is added, it becomes easy to collect the tracking devices and extract candidates.

The evaluation function used to calculate the coupling strength between the tracking devices in the candidate extraction processing is recorded in advance with the attribute values such as the position and speed of the tracking device, and the color and shape of the change area being tracked. The process may be replaced with a process of extracting a candidate for a moving object using knowledge on the shape, the movement form, and the like of the intruder that is running. As an example of such processing, a processing module is provided which replaces each information listed above between two tracking devices with a numerical value indicating whether the information is close (similar, tracking the same moving object) or not. In this case, the output results of these processing modules are simply added, or priorities are set for each information, and the weights added by weighting according to the priorities are used as the strength of the connection. Method. For example, by using information on the form of the intruder, the probability that the moving object candidate is truly an intruder is increased. Furthermore, without using what can be expressed as a numerical value called the coupling strength, it is determined whether or not each of the tracking devices can be included in the same candidate in each of the above processing modules, and the result is interpreted according to the priority order. Alternatively, a rule-based processing mode may be used. Further, the above-mentioned knowledge may not be recorded in the candidate extracting unit but may be recorded in advance in the knowledge providing unit 44 and provided in response to a request from the candidate extracting process. .

Each of the candidate lists created by the candidate extracting unit 42 does not consist only of a set of tracking devices, but also includes position information of the entire candidate and a color obtained from an area where the tracking devices are arranged. It may include attribute values such as shape information. A feature extraction process is newly added to the candidate extraction process, and at the same time as the candidates are extracted in the process, the above information is extracted from the region in the image where the candidates exist, and a candidate list is created by adding the information to the candidates. What should I do? If there is an additional attribute value in the candidate, the process of determining whether or not the attribute value is close to the process of associating the candidates performed by the result interpreting unit 43 is added.
The result is more reliable. Even if the above-described feature extraction process is not included in the candidate extraction process, it may be operated in parallel as another process, and the sequential process may be started in response to a request from the candidate extraction process.

In the above embodiment, the candidate extracting unit 42
The result interpreting unit 43 and the knowledge providing unit 44 operate simultaneously in the same host computer 4, but one or all of them may operate simultaneously in separate host computers prepared in advance. .
In this case, since the processing is divided by a plurality of computers, each processing can be executed quickly, and the tracking and result interpretation processing can be performed at a shorter time interval.
In addition to being able to more densely extract information about the intruding object, it is also possible to detect an intruding object that moves faster than when the division processing is performed.

[0057]

According to the present invention, a set of tracking elements connected by a virtual connection having an intensity equal to or greater than a predetermined value is extracted as a candidate for a moving object, and the moving path of the moving object is determined from these extraction results. Even if there are an unspecified number of intruding objects, such as humans, in the detection area, which are interpreted, the movements of multiple intruders can be detected by interpreting the results while tracking them. can do. In other words, even if there are multiple intruders in the surveillance area and they make complex movements, such as crossing each other, and it is very difficult to interpret the tracking result after the tracking is completed, A great effect in practical use can be obtained, for example, the detection of the moving route can be performed.

[Brief description of the drawings]

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

FIG. 2 is a view for explaining a display example on a display device used in the embodiment shown in FIG. 1;

FIG. 3 is a flowchart for explaining the operation of the embodiment shown in FIG. 1;

FIG. 4 is a flowchart for explaining the operation of the embodiment shown in FIG. 1;

FIG. 5 is a diagram for schematically explaining the operation of the embodiment shown in FIG. 1;

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... TV camera, 2 ... Change detection part, 21 ... Reference image memory, 22 ... Difference circuit, 23 ... Binary circuit, 24 ... Labeling circuit, 3 ... Tracking part, 31 ... Tracking device, 32 ... Control bus, 33 ... image bus, 4 ... host computer, 41 ...
Tracking device control unit, 42: candidate extraction unit, 43: result interpretation unit, 44: knowledge providing unit, 5, 6: display device, 7: A / D
Converter, 8 ... Display controller.

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06T 7/20 H04N 7/18

Claims (6)

(57) [Claims] A change detecting means for detecting a change area from a captured image; a tracking means having at least two tracking elements for tracking the change area detected by the change detection means; Candidate extraction means for obtaining a virtual coupling strength between the tracking elements and extracting a set of tracking elements having a coupling strength equal to or more than a predetermined value as a candidate for a moving object; interpreting the extraction result by the candidate extraction means; A moving object detection device, comprising: a result interpreting means for generating a moving path of an object candidate. 2. The method according to claim 1, wherein the candidate extracting means calculates a virtual coupling strength between the tracking elements using an evaluation function that takes a large value when the moving paths of the tracking elements are similar. The moving object detection device according to claim 1. 3. The method according to claim 1, wherein the candidate extracting means performs a process of extracting, as a moving object candidate, a set of tracking elements having a connection strength equal to or greater than a predetermined value at all times when the tracking means tracks the change area. Item 7. The moving object detection device according to Item 1. 4. A candidate list at each time point where a set of tracking elements is set as a moving object candidate by candidate extracting means, and the candidate list string is converted by a result interpreting means into a data string representing a moving path of the moving object. 2. The method according to claim 1, wherein
The moving object detection device according to any one of the preceding claims. 5. The moving object detecting apparatus according to claim 1, further comprising knowledge providing means for providing knowledge requested by the candidate extracting means and the result interpreting means. 6. The apparatus according to claim 1, wherein the result interpreting means performs a process of associating a candidate for a moving object that can be determined to be the same from the candidate extracting results at a plurality of consecutive times by the candidate extracting means.
The moving object detection device according to claim 1.