JPH04192781A - Tracking device - Google Patents

Abstract

PURPOSE:To surely execute the tracking of a change area by extracting the change area from a picked up image, setting an area including at least a part of the change area as a tracking area, obtaining the moving direction in the tracking area of the part of the change area and moving the tracking area. CONSTITUTION:When a coupling component possessing more than a prescribed area is detected in a labelling circuit 9, a host computer 5 directs in a tracking commanding part 10 that the tracking of the coupling component should be executed in a tracking part 4. A local module 13, which is connected with the computer 5 via a control bus 11, receives the command from the computer 5 via the control bus 11. In a tracking area setting part 14, the tracking area is set centering about a point O established desirably on the coupling component. The tracking is executed by consecutively moving the tracking area by using the moving vector obtained in the tracking area. The tracking area is outputted as a frame, for example, in a graphic display controller 18. The frame and an image inputted from a TV camera 1 are displayed simultaneously. Thus, the change area can be consecutively tracked without fail.

Description

DETAILED DESCRIPTION OF THE INVENTION [First aspect of the invention] (Industrial field of application) The present invention relates to a tracking device that detects and tracks a changing region in a captured image.

(Prior art) In recent years, surveillance systems (tracking devices) have been developed that use i''V cameras to capture images of surveillance areas and automatically detect intruders and objects entering the surveillance area. In order to detect an intruding object in such a system, the difference between images at short time intervals or the difference between a background image without an intruding object, etc. and the current image is calculated, and the changed part is detected from this difference image. 2゜methods have been used, however,
Simply detecting changes based on differences results in the detection of noise components due to environmental changes as well as intruding objects, and there is a problem in that it is difficult to distinguish between them. Noise components due to environmental changes include, for example, changes in brightness due to lighting changes.

Things that change in brightness within a certain area, such as the reflection of light on the water surface or the rustling of trees, appear as large differences in areas other than the invading object. Conventionally, it has not been possible to remove such noise components and reliably selectively detect the intruding object to be detected.

In addition, simply detecting a change based on difference does not necessarily result in a change area of -. in the difference image for an intruding object of -. There is a problem in that it is not possible to accurately remove noise components, integrate multiple change regions originating from one object, and determine the presence, number, trajectory, etc. of intruding objects.

Among the above problems, a solution called a block integration processing method (Japanese Unexamined Patent Publication No. 2-5997Ti) has been proposed regarding the problem of integrating a plurality of change regions J originating from one object. As shown in Fig. 9, when an intruder is divided into multiple changed regions such as the head, torso, and legs in the difference image, between the changed regions extracted at two consecutive points in time, Find the corresponding change area based on the features t18 such as area and shape,
This method calculates the movement vectors between corresponding change areas and determines change areas with similar movement hectares as originating from the same object. This method works effectively if the object is always divided into the same components. However, from a shooting perspective, it is often not fixed in which part of the object the changing region that is differentially extracted is located. The reason for this will be explained below. First, the way brightness changes due to illumination changes differs depending on the part of the target object. Furthermore, if the target object is a three-dimensional object with depth or unevenness, the appearance (area, shape, etc.) changes depending on the orientation of the object. Also, even if the target object is not a completely rigid body (such as a person wearing clothes), the shape of the object changes at each point in time. When a part of the background and a part of the target object have the same color or the same brightness, the same color or the same brightness part will not be extracted as a changing area while the target object passes the background with the same color or the same brightness. And so on. As described above, in situations where it cannot be said that a corresponding change area is always detected at each point in time, this block integration processing method is preferable because there is no corresponding change area and the moving hectare cannot be calculated. It may not work. If such a malfunction or non-operation occurs, it will be extremely problematic for the monitoring system 11 (tracking device). .

(Problem to be solved by the invention) As described above, in conventional tracking devices, it is not possible to eliminate the influence of phenomena that cause changes in brightness other than the target intruding object, resulting in erroneous detection. Furthermore, there is a problem in that when an object marks a plurality of changing regions, it is difficult to determine that the objects are the same object.

The present invention was made in consideration of these circumstances, and its 1st purpose is to remove noise components caused by environmental changes, integrate multiple change areas originating from objects, and eliminate intrusion. In order to determine the number and movement of objects, "ljf'
To provide a tracking device with excellent environmental resistance, which allows even a situation expert who is not necessarily able to obtain +6 of the change area to reliably track the change area.

[Structure of the invention]

(Means for determining the problem fM′) A tracking device according to the present invention extracts a changed region from a captured image, sets a region including at least a part of the extracted changed region as a tracking region, and tracks This method is characterized in that the changing area is tracked 71 by determining the moving direction within the tracking area of a portion of the tracking area and moving the tracking area using this moving direction.

In addition, the captured image is displayed,
(Display on 7RT, etc., extract the change area from the captured image, set a tracking area that includes at least a part of the extracted change area, display it on the display β stage, This is characterized in that the display within the tracking area is moved at a rate of -I]J= as the tracking area moves.

(Operation) According to the present invention, even if a single object is extracted as a change region that has a plurality of unspecified parts, and the way of separation is different at each point in time, the change continues. By tracking the area, noise components due to environmental changes can be removed since tracking them does not result in meaningful movement, and if the tracking paths of multiple changing areas are similar, it can be determined that they are the same object. I can make judgments.

(Example) Hereinafter, a tracking device according to an example of the present invention will be described with reference to the drawings.

Figure 2 is a schematic configuration diagram of the tracking device. This tracking device f
Broadly speaking, f includes a TV camera 1 that captures an image, a change detection unit 3 that extracts a changed area of the captured image, a host 1 to a computer 5 that outputs a tracking command for a darkened area, etc.
It consists of a tracker that tracks the area of change. To explain in detail,'
The image input from the J'V camera 1 is digitized by the A/1 converter 2. The change detection section 3 is -
The difference between the current image taken as described above and the image stored in the reference image memory (3) is determined by the difference circuit 7.The reference image memory 6 contains the background image at a time when there is no target object. 1 and store it.The difference result is binarized by a binarization circuit 8 using an appropriate threshold value to determine whether or not it is a changed part.This allows extraction of parts that have changed significantly with respect to the reference image. As a result of binarization, the labeling circuit 9 gives different labels to the components connected as change regions in the difference image.The labeling circuit 9 also applies the area, centroid position, and circumscribed rectangle for each connected component. (Hereinafter, ``direct components'' will be used as a synonym for ``variable area.'') The computers 1 to 5 are not in the form of workstations, personal computers, or such independent computers; This is realized by a processor.The host 1-computer 5 controls the operation of the entire device.The rough 1-ware on it also controls the tracking result determination unit 1.
When a connected component with a fixed area of 7 to 7 is detected by the labeling circuit 9, the tracking command unit 0 instructs the tracking unit 4 to trace the connected component. order to do so. In this embodiment, the tracking unit 4 is composed of a multiprocessor. The individual processor parts of the multiprocessor, here local modular modules, keep track of a connected component. When the number of local modules is small, one local module can track multiple connected components.

The local module 13 is connected to the computer 5 to the host 1 by a control bus 11, and receives commands from the computer 5 through the bus 11. Further, the processing results are transmitted to the host computer 5. The local module 13 can input any part of a moving image from the A/D converter 2 through the image bus 12.

FIG. 2 shows the configuration of one local module J3. When the position (address) of the part of the image to be captured is set in the registers of the window 1, roller 1, and roller 1.3a, that part of the image flowing on the image bus 12 can be input to the window memory 3b. Local processor +3c processes this input image. Setting of the registers of the window controller 1 to roller 3a can be done from both the host computer and the local processor 3c. This switching is performed by the bus interface 1.3d. When the host 1-computer is set, the host interface 13d connects the control bus 11 and the local bus 13e. When local processor 3c performs settings, bus interface +3
d separates the control bus 11 from the local bus 13e. The plurality of local modules 13-1 to 13-D can operate independently and in parallel.

The operation of the local module] 3 as the tracking unit 4 will be explained below. Here, consider the case where one local module tracks one connected component.

First, in the tracking area setting unit 14, a rectangular area of an appropriate size containing the circumscribed rectangle of the connected component is determined, and its position (address) is stored in the register of the window controller 1-roller 3a. .

This may be determined by the host 1-computer 5 and performed from there, or information regarding connected components may be sent to the local module, and the local module may determine and set it. Then, the image of that part at the current time (time 1.) is stored on the image bus 12 (window memory 31).
Import from. Next, the image of the same portion at the next time point (time 1.+Δt) is also taken into the window memory], 3b.

Then, a tracking area is set around a point O (for example, the center of gravity) that is appropriately determined on the second connected component.

Next, in the movement direction detection small area setting section 15, the third
As shown in the figure, a small area (
For example, set ro-r8). For example, the size of these × areas is 512151 for the entire image.
When the size is 2 pixels, ux = Wy = 50 pixels, 5x = sy = 1. x = 1. Set to approximately y = ] - 0 pixels. Alternatively, the sizes of these regions may be appropriately varied based on the area of the connected components or the size of the circumscribed rectangle. Also, here we set 3×3=9 small areas, but if n and m are arbitrary integers, n×m=
You may set n m small areas. For each small region r, d (1)=Vj, (t +Δt;) −v
](t,). Here, Vi(t, +△12)
is time 1. The average value of the density of pixels in the small area at +Δt, Vj(t), is the same value at time t. However, if the value obtained using this formula is less than or equal to a predetermined threshold value (th), then dl)=O.

Using this d(j), in the movement vector 1-le calculation unit 1[j, the movement vector I-le (u+v) is calculated by the following formula.
:do.

(However, Σ can be generalized as Σ) i, = oi, = n Here, Xi, + y are the coordinates of the center of the small area j when the point ○ is the origin. Further, f is a coefficient and is appropriately determined depending on the application example. f is determined by the difference in brightness between the background and the object and the speed of movement of the object on the image, and can be set to 0.5 depending on the application.
Set it to about 2. By finding the moving bags 1 to 1 in this way, the tracking of ]-frame 11 is completed.

To track the next frame 11, the tracking area setting unit 14 again refers to the movement backgrounds 1 to 1 of the previous frame to determine a rectangular area for inputting a new image, and then determines the rectangular area for new image input.
, 3b also shows the image of that part at time t+Δ11 and 1, +2
The process begins by capturing an image of the same portion at time ΔL. Then, the center ○ of the tracking area R in the previous frame 11
is moved by the movement vector obtained in the previous frame to set a new tracking area. Thereafter, new moving bags 1 to 1 are obtained in the same manner.

As described above, the moving object 1-
Tracking is performed by moving the tracking area one after another using the tool.

The operation of the tracking unit 4 described below-1 can be visualized on the display device J9. Outputting the tracking area set in the tracking area setting unit 14 as a frame, for example, on the graphic display controller 1-roller 18;
The frame representing this tracking area and the image input from {rV force=12-mera} are displayed at the same time.

FIG. 4 shows an example of the screen of the display device 19 at this time. However, this figure shows screens at two different times on the same screen for convenience in the following explanation. A detailed explanation will be provided later.

As described above, the tracking device according to this embodiment can reliably and continuously track the changing area. Then, from this tracking result, the tracking result determination unit 17, which will be described later, distinguishes noise components due to environmental changes from the target object, or distinguishes noise components caused by environmental changes from the target object, or when multiple change regions originate from the same object. You can judge and track movements.

At this time, several methods can be used to establish the operational relationship between the change detection section 3, the tracking section 4, and the tracking result determination section 17 depending on the intended use and application.

First, when only one target object enters the screen, or when only one object needs to be detected, the most common method can be used. In this case, change detection is performed, and when a changed area is extracted, change detection is stopped and only tracking processing is performed. In this case, even if there is only one object, there may be multiple change regions for it.

For most purposes, only a suitable one of them, for example the one with the largest area, is tracked. When a multiprocessor system is used as in the embodiment, multiple change areas are tracked by different local modules.

In the local module tracking process, the movement vector [・
In the case of 2 or 0, it can be considered that tracking has failed, the object has stopped, or the extracted change region is not derived from a moving object. When this happens, the tracking process is stopped and change detection starts again. Note that depending on the nature of the object, tracking may not be stopped immediately even if the movement vector becomes O, but tracking may be stopped when the movement vector is O at several points in time.

When there is only one tracking result-specific path object moving from computer 5 to host 1, if tracking continues for a certain period of time, it is determined that the object is moving.

If the object stops midway, tracking by the tracking unit 4 will temporarily stop. However, when the object starts moving again, a change will occur, and the change detection unit 3 will detect it.
The chase begins again.

Therefore, even if tracking is interrupted midway, if tracking starts again after a while from near the interrupted position, this is considered as one movement and the movement of the object is determined.

Even when tracking multiple change areas, in this case only one object is detected, so if the moving bag 1 cannot be detected during the process, the tracking is stopped and one of the change areas is detected. If the object can be tracked continuously enough to be determined to be the movement of the object, it is assumed that the objective has been achieved. If tracking of all changed areas fails, change detection is started again.

Next, a general case where the number of objects is not limited to one will be explained. In this case, it is necessary to constantly move the change detection section 3 or perform more complicated processing on the tracking result determination section 17.

First, the change detection section 3 is kept in constant motion. Once a region of change is extracted, local modules] 3 are assigned here and there to track it. However, if there is an object in the image, there will be a constant change area in the difference result, so you need to make sure that you do not track a new part that is already being tracked by another local module. No.

This can be done by checking whether there are any tracking results near the changed region when it is extracted/extracted. After receiving the center of gravity and the position of the circumscribed rectangle of the changed area from the change detection unit 3, the computers 1 to 5 check the tracking results of the local module that is currently executing the tracking process, and find the tracking closest to the position of the changed area. If there is a change area, tracking processing is not activated for that changed area.

In this part, the correctness of the tracking process may be further confirmed as follows. If the tracking process is correct, there should be a changed area at the location where the moving background is tracked. Therefore, when the local module that is tracking calculates the moving bag 1 to 1, it sends 1N information to the host 1 to computer 5, and confirms whether the position is within the circumscribed rectangle of a certain change area or whether the area around the position is Check whether the result of the binarization circuit 8 is a change area. If these conditions are satisfied, it is assumed that the tracking is correct and the tracking process continues. If not, it is assumed that the tracking process has failed and the tracking is stopped. In this case as well, the tracking may be stopped when such a phenomenon occurs a plurality of times, instead of making a determination based on the results of - times. Including such processing will make tracking processing more reliable.

The tracking result determination process by the host computer 5 is performed as follows. If the target object is reliably extracted by the change detection unit 3 as one change area or multiple change areas that are always in a constant positional relationship,
The determination process may be the same as in the case of one object. This can be considered in applications where the contrast between the object and the background is high, and the object is a planar rigid body. In such a case, if the individual tracking results are repeated to some extent continuously, it may be determined that there is an object that moves in that way. Even when an object is divided into a plurality of changing regions, their mutual positional relationship remains constant, so it is only necessary to combine the parts that maintain such a relationship based on the individual tracking results and determine them as one object.

Below, a description will be given of tracking result determination processing in a general case where the following two conditions cannot be assumed. In the general case,
For example, something like the one shown in Figure 4 occurs.

In Figure 4, the change area for one object is 40°41
, 42, some of which are 41°42
Tracking may be interrupted because no change is detected, or different parts (43, 44) of the object may be detected as change areas and tracked. In addition, there are cases in which a changed region has been generated once but can hardly be tracked (45), and tracking results derived from other objects (46, /17). The tracking device according to this embodiment is especially necessary in cases such as the one shown in FIG.

FIG. 5 shows the tracking result determination unit 17 of the host 1 to computer 5.
Flowchart 1 of the tracking result determination process to be performed is shown. Divide the directions in the image into an appropriate number (for example, 16 directions)
Then, prepare a direction-specific chiffle that can store the tracking results for each direction.

This table is reset at the start of the device (step 501).

This process constantly receives and receives intermediate results of tracking (movement bags 1 to 1 determined for each frame 11) from each local module 13 (step 502). If one local module continues tracking for a predetermined length, the process goes to step 511 (step 504'/es).

If not (No in step 504), the moving bag 1 to 1 during the predetermined time is calculated (step 505). This can be done by using the movement vector obtained in each frame 11 as is, but it can also be used when the movement of the object on the image is slow.
Generally, the sum of movement vectors for several frames 11 is calculated. The sum of the vectors for a predetermined number of frames may be calculated, or the movement vectors may be continuously added, and when a predetermined size is exceeded, the movement between them may be calculated. This process is performed for the purpose of smoothing, since if the movement is too short, the moving background may not be stable. The moving vector I-1 obtained in this way is written into a direction-specific table for the direction of the vector I-1 (step 506). At this time, record the time as well as the position and size of Muku 1 Hell. In this program, data after a certain period of time is deleted from the direction-specific table (step 507). Therefore, it was almost impossible to trace 45
are deleted from the table after a predetermined period of time and do not adversely affect object detection.

Incidentally, if the tracking cannot be performed at all, the local module informs the host 1 to the computer 5 that the tracking has failed, and is not involved in this part of the processing.

Furthermore, when a method of accumulating movement vectors is adopted, vectors that do not reach a predetermined size within a certain period of time are also not written into the direction-specific table.

Next, it is checked whether a predetermined number of data similar to -1- has been stored in the direction in which it was written (step 508). If not, the next point in time tracking information is accepted and the process returns (step 508, No). If Yes, check the position and time of the moving bag 1 to 1 in that direction (in some cases, you may also check neighboring directions),
The parts forming one group are found (step 509). If the information is not organized, the process returns to receiving the tracking information (No in step 509). If a cluster is detected, it means that one object has been detected (step 51(1)).

From the field of view, given tracking length, and number of moving vectors,
This is often sufficient, but in applications where the tracked length is relatively small within the screen, the object may have already been detected. Therefore, it is checked whether there is any result that is continuous with the current result among the already detected results, and if there is, it is determined that it is the same object (step 51 (1)).

When finding an object that crosses the screen (field of view), if the movement vector approaches the periphery of the screen in steps 504 and 508, the object should move in the direction of Yes.

If Step 504 is Yes, it is certain that there is an object at this point, so check the data for directions similar to the moving direction of the object in the table by direction, and select those whose position 11q and time form a group (Step 5 ]], ),
, and it is determined that an object has been detected (step 51(1)).

When an object is detected, the moving pattern data based on the detected object is deleted from the direction table (step 51).
2). Then, after receiving the tracking information, the user returns to the destination.

Note that the above-mentioned judgment method can be used in cases where there are limitations on the object, etc. and a simple judgment method can be used (in cases where there is only one object as mentioned above, or where there is only one or multiple objects). Of course, it can also be used in cases where the variable area is always extracted as a change area having a fixed positional relationship.

An example of the embodiment has been described above, but various modifications and additions can be made to it will be explained below.

In the second embodiment, the change detection part is implemented by a dedicated circuit, the tracking part by a multiprocessor system, and the tracking result determination part by a host computer. This enables high-speed processing and can handle cases where the target object moves by a large angle. However, depending on the application, all or some of them may be one S P (Djgj, t, al! SjV, nalPr
occssor) and dedicated circuits, but other functions may be performed by a microprocessor or other digital computer.

First, a modification regarding the change detection section 3 will be described. .

-1] In the embodiment described above, the changing area is extracted by the difference with the reference image that has been prepared in advance, but this is the differential (difference) between the consecutive 2 or more 1- images. It may also be done by calculation. In this case, the possibility of erroneously extracting a changed area due to environmental changes is reduced. However, in this case, since the outline of the object is detected, it is difficult to form a unified connected component. Therefore, it is necessary to set the tracking area by dividing the appropriately extracted change area. Furthermore, it may be better to appropriately determine the center of the tracking area rather than the center of gravity of the connected components.

Even when a reference image is used, the reference image may be recreated at appropriate intervals in order to prevent erroneously detecting areas of change due to environmental changes.

In the embodiment described in 1-1, the difference image is divided into 2 with a fixed threshold value.
The area of change is extracted by converting it into a value. Changes can be detected more reliably by varying this threshold value depending on the density of each part of the image.

Furthermore, when considering changes, only the brightness of the image is considered in the embodiment described in 1-1, but detection based on other changes such as changes in color or texture may be used independently depending on the application. . For example, if it is known that the object has a specific color, the change in color can be used to reliably detect the change. Alternatively, a combination of multiple change detections may be used. In this case, depending on the target,
There are cases in which changes can be successfully detected with a certain process. In such a case, the results can be communicated to the tracking processing, and the tracking processing can also track the information by 1%fj to improve the reliability of the processing. It is Noh.

In the second embodiment, which of the changing regions to track is determined based solely on the size of the area of the region. If the shape information is also taken into account when selecting, the target can be selected more reliably depending on the application. Furthermore, since the outline of a stationary object tends to appear as a differentially changed part, it is also possible to add processing such as removing such thin edges or ignoring changes that appear at a certain position in the image.

Next, a modification regarding the tracking section 4 will be described. The tracking method in the embodiment described in L is to determine the moving direction 2 from the temporal change in concentration around one point. .

This has the advantage of high-speed processing and reliable operation in terms of tracking moving objects.

However, since we are not looking at the properties of the object, when other moving objects come around the object, their) 11! Sometimes the direction of movement is output incorrectly 3, l”, -, 1
':1. The following tracking method can also solve this problem. This is because the continuous movement of the moth [in the process of determining 1 trace and fruit, moving Muk 1 Hel to the directional table at the A111 size of A111 of 1 to several frames 13 - old, from there -] This is because the trajectory of each object can be comprehensively determined from the 11 divided small steps and a series of large movements.

Now, using another tracking method that looks at the properties of the object,
You can solve two sets of problems. For example, in Fig. 3, the correlation coefficient between the current image of the small area ro, which is the center of the tracking area R, and the next image of all points in a certain range around ro is calculated, and the maximum correlation is calculated. Tracking may be performed in the indicated direction. Or, to speed up this, for example r1 to r
It is also possible to use a method in which correlations are found for the 8 points of 8, and the areas around the larger ones are examined in detail. In addition, as a simplification of the above method, in order to examine the similarity between the small region r・0 and surrounding small regions, instead of the correlation, the absolute value of the difference between the pixel values at the corresponding position It is also possible to calculate the sum and use the sum. Or 5SD, which is a fast calculation method of this
A (Sequut, i, al Sjmj, ], ar
jtyDetectjon Allgorjthm) may be used. 5SDA is a process of finding the minimum position of the sum of the absolute values of the differences between the small areas to be compared one after another. If you find the values and add them, and the added value exceeds the minimum value of the sum so far, the sum of the differences for this small area will be greater than the minimum value, so from now on, it is meaningless to add. Therefore, the method is to abort the calculation and proceed to the next small area. In such a case, the properties of the object are being looked at, so the probability of tracking the same object is high. However, highly correlated positions are often not clearly determined, and tracking often fails.

In addition to correlation, find feature points such as edges and corners,
A method of tracking by continuously determining the position can also be used. This method is highly reliable if those feature points are clearly determined. However, this method does not work well for objects for which it is difficult to obtain these feature points.

The tracking method most suitable for the application may be selected and used, such as the following. Alternatively, you can use all your skills and eventually adopt the ones that can track them well. or.

All tracking results may be sent to a tracking result determination process and the detection of meaningful movements may be left to that process.

In this way, performing multiple processes requires a large amount of computing power, but the reliability of the processes is improved.

Furthermore, the tracking method used in the embodiment described in -1: can be widely applied by changing the tracking characteristics of the local module (3).

For example, by changing the sign of the value of f, the 1 in particular for the object changes. If f is positive, the characteristic of following the object can be realized, and conversely, if f is negative, the characteristic of trying to move away from the object can be realized. In addition, as shown in the following equation, the moving pixel (u
, v) by φ, it is also possible to make the module track in a direction shifted by an angle φ with respect to the movement of the object, as shown in FIG.

By using these characteristics in combination, it becomes possible for each module to selectively track the front edge, rear edge, front diagonal edge, and rear diagonal edge of an object as shown in FIG. Note that the double method can be replaced by introducing a weight W(]) into dO) for each small region and changing its sign and absolute value, as shown in the following equation.

(However, Σ can be combined with Σ.) j = oi, = 0 In other words, by using this weight W(1), for example, it is possible to predict the direction of an object and track it based on the moving back of the engine. It becomes possible.

Furthermore, as an application of the above-described method, by arranging a large number of modules as shown in FIG. 8, the axis contour of the object can be extracted. At this time, the host computer controls each module so that the modules do not track the same location and to prevent the modules from moving away from the object due to disturbance.

Alternatively, local modules communicate with each other so as to act as restraints and restraints on each other.

Compared to conventional methods such as differential processing and binarization, this contour extraction method uses motion information from a large number of modules, so it is less affected by the environment and can accurately extract contours. Furthermore, it is possible to capture/ML the movement at the same time as contour extraction. From this contour information and its movement information, we compare it with the shape pattern and movement pattern of the object that we have in advance,
Objects can be recognized.

Further, the series of methods described so far can be realized on one local module, which functions as a module that tracks an object and extracts its outline.

〔Effect of the invention〕

As explained above, according to the present invention, even in cases where correspondence between changing areas cannot be established, changing areas can be reliably tracked, noise components caused by environmental changes can be removed, and noise components caused by environmental changes can be removed.
Great practical effects can be achieved, such as the ability to realize a tracking device with excellent environmental resistance, which makes it possible to integrate a plurality of changing regions originating from one object and observe its movement.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram of an embodiment of the present invention, FIG. 2 is a configuration diagram of the local module 13, FIG. 3 is an explanatory diagram of a tracking area and a small area for detecting moving direction, and FIG. 4 is an illustration of one object. FIG. 5 is a diagram showing an example of a case where the tracking result is tracked separately into a plurality of change regions, and FIG.
The figure is an explanatory diagram of tracking in directions different by an angle φ, Fig. 7 is an explanatory diagram of selective tracking of object edges, Fig. 8 is an explanatory diagram of object contour extraction, and Fig. 9 is an explanatory diagram of a conventional example. be. I TV camera 2 ・A/1 converter 3・
- Change detection unit 4... Tracking unit 5... Post computer 6... Reference image memory 7... Differential circuit
8. Binarization circuit 9...Label link circuit
]0 Tracking command part 1]・Control lotus 12
Image buses 13-1 to n, Local module 14, Tracking area setting unit 15, Moving direction detection small area setting unit 16, Movement vector) ~ Lumi 1 calculation unit 17, Tracking result determination unit] 8 Graphic display controller 1 ~ Laura 19・
・Display device 13a ・Window controller 1 to roller 1.3b Window memory 13c Local processor], 3d Path interface agent Patent attorney Noriyuki Chika Figure 8 Figure 9

Claims (2)

[Claims] (1) A change detection means for extracting a change area from a captured image, a setting means for setting an area including at least a part of the change area extracted by the change detection means as a tracking area, and a setting means for a moving direction detecting means for determining a moving direction within the tracking area of the set portion of the tracking area; and a movement for moving the tracking area based on the moving direction determined by the moving direction detecting means. A tracking device, comprising: a means for tracking the changed region by moving the tracking region by the moving means. (2) A display means for displaying a captured image, a change detection means for extracting a changed area from the captured image, and a tracking area that includes at least a part of the changed area extracted by the change detection means. A tracking device comprising: means for setting and displaying on said display means; and tracking means for moving a tracking area displayed by said means in accordance with movement of said changing area.