JPH0894729A - Target classification method - Google Patents

Abstract

(57) [Abstract] [Purpose] A target classification method used in an image processing system that tracks a minute target extracted from a digital image input through a TV camera. The objective is to reduce and enable continuous scan-to-scan tracking. A target having a brightness level equal to or higher than a first threshold value is extracted from the targets having a high brightness level extracted from the digital image, and a feature amount including the brightness level of the extracted target is calculated and stored. Then, of the stored brightness levels at the current extraction, a target having a brightness level equal to or higher than the second threshold value is determined to be a classification target to be tracked between scans, but the brightness level at the next extraction is the second level. When it is lower than the threshold value, if it is extracted, it is regarded as the same classification target, the feature amounts of the previous time and this time are compared, and if the difference between the feature amounts is within the setting range, the two classification targets are identified as the same target. Then, it is configured so that it can be tracked.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a target classification method used in an image processing system for tracking a minute target extracted from a digital image input through a television camera.

For example, when extracting and tracking a target from a digital image input through an infrared television camera,
When the target extracted in the previous scan is not extracted in the current scan and is extracted in the next scan, it may be determined as a new target and the tracking between scans may not be continued.

Therefore, it is necessary to reduce the omission of extraction of a target having a large fluctuation in the brightness level to enable continuous tracking between scans.

[0004]

2. Description of the Related Art FIG. 4 is an explanatory diagram of digital image generation, and FIG.
6 is a schematic diagram of a main part of the target classification means, FIG. 6 is an explanatory view of a conventional target classification method, FIG. 7 is a schematic view of a filter, (a) is an example of a preprocessing filter, (b) is an example of an extraction filter is there.

The description of FIGS. 4 to 7 will be given below. In FIG. 4, the infrared detector 12 is provided with a plurality of photodiodes that photoelectrically convert incident infrared power and output charges.

Therefore, the infrared detector 12 extracts the electric charge corresponding to the infrared power from the infrared light incident through the optical system 11. The electric charge is converted into a series of current signals by a multiplexer (not shown), and then a voltage signal is obtained. Converted to A /
It is added to the D converter 13.

The A / D converter converts a voltage signal into a digital signal (
Digital image) and offset / sensitivity adjustment part
It is sent to the target classification processing means via 14. As shown in FIG. 5, the target classification processing means is a program ROM that stores a program for causing the CPU to execute the CPU 21, the processing flow shown in FIG. 6 (consisting of steps S25 to S27).
It has 22, RAM 25 and interface portions 23, 24 for interfacing with external devices, and the above digital image is stored in the RAM 25 via the interface portion 24.

Therefore, the CPU 21 reads the stored digital image according to the processing flow shown in FIG. 6 and applies the preprocessing filter S 25 to extract a minute target having a bright brightness level in this image, which is all soft. To process.

For this reason, as shown in FIG. 7 (a), for example, 3 × 3 pixels in the inner frame ("IN" in the figure is called the IN side), 5 × 5 pixels in the outer frame (in the figure) The "OUT" part is called the OUT side) is applied to each pixel sequentially from the edge of the digital image, and the difference between the sum of the IN side brightness level and the OUT side brightness level (ΣIN side −ΣOUT Side) is replaced with a new image, but it is executed for the entire one screen (this processing is preprocessing).

As a result, the brightness level is high (bright).
Since only the part is extracted, this filter has the function of making the part on the IN side stand out from the part on the OUT side. Also, when there is a bright large target exceeding 5 pixels × 5 pixels, the values of (ΣIN side−ΣOUT side) are the same, and if they are different, it becomes black and erased on the pre-processed screen.

However, since there is a difference in brightness level between the inside and outside (background) of the boundary at this target boundary, this part remains. The size of the above frame is determined by software.

That is, by applying a pre-processing filter to the input digital image, a minute target having a high brightness level can be taken out. Now, the extraction filter shown in Fig.6-S26 is further applied to the digital image to which the pre-processing filter has been applied. This filter has an inner frame (IN) and a middle frame as shown in Fig. 7 (b). There are three frames, (MID) and outer frame (OUT), but in the conventional example, the middle frame is not used, and (ΣIN side −ΣOUT
Value) and compare the background on the IN side with the background on the OUT side.

Here, as shown in FIG. 6, a target threshold value is applied to the extraction filter. This threshold value is the ratio (S) of the difference between the background and the target brightness level in the input digital image.
/ N), but set higher than the highest brightness level of the background with noise.

For example, when the target brightness level extracted through the preprocessing filter and the extraction filter is 15 to 20, the target extraction threshold value is set to 7, which is about half, for example. Now,
Using a filter with three frames for the digital image obtained by applying the preprocessing filter (ΣIN side −ΣOUT
Side), the minute target whose difference has a brightness level equal to or higher than the target extraction threshold value is extracted, and the feature amount (for example, target level, target size, target position) is calculated.

Here, the target level is a value of a brightness level equal to or higher than a target extraction threshold value, the target size is a size of a pixel entering the IN side, a value between a minimum of 1 pixel and a maximum of 9 pixels, and a target position is an image. It is the information indicating the direction of the position of the target inside and the direction of the image itself.

Then, the calculated feature amount in the current scan is stored in the target data storage unit 251 of the current scan, but the target data in the previous scan is stored in the target data storage unit 252 of the previous scan.

Therefore, in the target identifying section of FIG. 6-S27, the identification processing is performed by comparing the feature amounts stored in the two storage sections, and the minute target extracted in the previous scan and the minute target extracted in the current scan are extracted. Whether or not the minute targets are the same target is identified, and the feature amount of the current scan is stored in the target data storage unit 252 of the previous scan for the next comparison.

For identification, the state of the current scan (positional state) is estimated from the tendency of the previous scan, and if the calculated position is almost the same as the estimated position, the same target is identified. Also compare the levels.

When the determination is completed, the process returns to the preprocessing filter in S25 and the above processing is repeated.

[0020]

FIG. 8 is a diagram for explaining the cause of fluctuations in the brightness level (in the case of two elements). As described above, in the preprocessed digital image, not only small targets but also the edges of large targets (ships, etc.) may be extracted, so the brightness level of the edge background is cut and only small targets are extracted. In order to extract, it was necessary to set the target extraction threshold of the extraction filter high.

However, the signal level of the extracted minute target
The (signal level output by the detection element) is not always constant due to, for example, variations in element characteristics or disturbance, so when the signal level becomes low, it may fall below the target extraction threshold value.

Here, as an example of the cause of the fluctuation of the brightness level, there are variations in the sensitivity of the sensing element and differences in the target sensing position. In FIG. 8, ~ in the figure is the number of scans. The image sensing element is not a single element, but a plurality of sensing elements are often used for imaging. When the sensitivities of the A element (odd number column) and the B element (even element) are different, the level detected by the element is different even if a target of a constant size is imaged with a constant radiance.

For example, at the detection position shown in the drawing, since the target is included in one element, the sensitivity of the element does not change and the respective brightness levels are the same. However, when two pixels are straddled as described above, both sides do not necessarily have the same sensitivity, and therefore the brightness levels are different.

Therefore, the minute target extracted in the previous scan is not extracted in the current scan, and when it is extracted in the next scan, the originally same target should be judged as a new target and the scan tracking is performed. May not be continued.

It is an object of the present invention to reduce extraction omission of a target having a large variation in brightness level and to enable continuous scanning-to-scan tracking.

[0026]

According to the present invention, a target having a brightness level equal to or higher than a first threshold value is extracted from targets having a high brightness level extracted from a digital image, and the brightness level is extracted for the extracted target. A feature amount including the feature value is calculated and stored. Then, of the stored brightness levels at the time of this extraction, a target having a brightness level equal to or higher than the second threshold value is determined as a classification target to be tracked during scanning. However, when the brightness level of the classified target determined this time becomes lower than the second threshold value at the time of the next extraction, it is regarded as the same classified target if it is extracted, and the feature amounts of the previous time and this time are compared, If the difference between the amounts is within the set range, the two classification targets are identified as the same target and can be tracked.

[0027]

FIG. 1 is a diagram for explaining the principle of the present invention. In the conventional target extraction method, the extraction condition is that the luminance level of the minute target has a value exceeding the target extraction threshold value (7 in the above).

In the present invention, the target extraction threshold value of the extraction filter S 22 in FIG. 1 can be made lower than that of the prior art (for example, 3) to cope with the fluctuation of the luminance level of the minute target, and the minute target can be extracted. It has become possible to reduce this.

Further, in the present invention, the conventional target extraction threshold value is set as the target classification start threshold value (eg, 7) of the classification target determination processing (S23), and minute targets having a brightness level higher than that are classified. Have the same goal, that is, a categorical goal.

However, even if the brightness level falls below the target classification start threshold value in the next scan due to a sudden level change, if it is extracted by the extraction filter S 22, it is regarded as the same classification target, and the target identification section Added support for tracking during scans.

In this case, it is necessary that the difference between the feature amounts at the current scan and the next scan is within the allowable range. As a result, it is possible to reduce omission of extraction of a target having a large fluctuation in the brightness level, and it is possible to continuously perform tracking between scans.

[0032]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 2 is an explanatory view of essential parts of an embodiment of the present invention, and FIG. 3 is an operation explanatory view of FIG. Here, the same reference numerals denote the same objects throughout the drawings. Further, the parts described in detail above will be briefly described, and the parts of the present invention will be described in detail. The operation of FIG. 2 will be described below with reference to FIG.

In order to carry out using software, as explained above, the program for executing the procedure indicated by the dotted line in FIG. 2 may be stored in the program ROM of FIG. 4 and the CPU may be controlled according to the program. .

In FIG. 2, the image acquired by the infrared television camera 15 is digitized by an analog / digital (A / D) converter and sent to a preprocessing filter S 21. The preprocessing filter S 21 filters the input digital image to extract a minute target having a high brightness level (bright) in the digital image and sends it to the extraction filter.

The extraction filter S 22 extracts a minute target having a level equal to or higher than the target extraction threshold value from the digital image obtained by applying the preprocessing filter, and calculates the feature amount of the extracted minute target. .

Here, the target extraction threshold value is the ratio (S / N) of the brightness level difference between the background and the target in the input digital image.
). Also, since this value changes depending on the digital image and is set lower than the conventional value, the value obtained by subtracting the maximum noise level from the target brightness level, that is, the value that seems to be the target, must be a value that can be extracted (for example, 3) .

For example, if the extracted target brightness level is 15 to
If the noise is 0 to 1 at 20, the threshold value is set to about 3. As a result, all targets having a brightness of 3 or more can be extracted by this extraction filter.

Further, the feature amount (target level, target size, target position) of the extracted minute target is calculated. Then, the target data storage unit 251 stores the feature amount of the minute target calculated during the current scan, and the target data storage unit 252 stores the feature amount of the minute target calculated during the previous scan.

The classification target determination processing unit S23 is higher than the target classification start threshold value added by the operation unit 34 with respect to the small target for the current scan stored in the target data storage unit 251. The minute target having the brightness level is determined as the type target for starting the tracking between scans, and the corresponding feature amount is sent to the target identifying unit S 24.

The classification start threshold is the extraction filter S
Among the targets extracted in 22 that are not small targets (for example, large targets such as ships above)
To a value (eg, 7) that can be removed.

The target identifying unit S 24 is a target data storage unit 252.
Identification processing is performed by comparing the feature amount at the time of the previous scan read from the feature amount with the feature amount sent from the classification target determination processing unit S23, and it is determined whether or not they are the same target.

As a result, even if the brightness level falls below the target classification start threshold value in the next scan due to a sudden level change, if it is extracted by the extraction filter S 22, it is regarded as the same classification target, and the target identification unit It was made possible to track between scans with S 24 (described later using FIG. 3). In this case, it is necessary that the difference between the feature amounts of the current scan and the next scan is within the allowable range.

Then, the target identifying section S 24 stores the feature quantity of this time as the previous feature quantity in the target data storage section 252 and sends the identification result to the display unit 33 via the target classifying section 32.

Now, FIG. 3 compares the case of performing the inter-scan tracking of the conventional example and the case of performing the inter-scan tracking of the present invention. Here, the target classification start threshold value (conventional target extraction threshold value) is 7, the target extraction threshold value is 3, and the double threshold setting extraction method is the name of the extraction method of the present invention. The horizontal axis shows scan numbers 1 to 13. (1) In the case of the conventional example Since the target threshold value in the classification target determination processing unit is the same value as the target extraction threshold value, a minute target having a brightness level exceeding the target extraction threshold value is extracted for the first time in 7 scans. Since this is done, the target number 1 is given to this minute target.

However, in 9 scans, the level is below the target extraction threshold value, so that it is not extracted, and when it is extracted again in 10 scans, it is judged as a new minute target and the target number is 2.
Is given. Since the target number is newly added, it is not determined that the target is the same target as the target number 1, and the tracking between scans is interrupted in 8 scans. (2) Case of the present invention In the present invention, the extraction filter S 22 performs target extraction with a threshold value 3 lower than that in the conventional example. Although the threshold value is exceeded in four scans, the threshold value is not exceeded at this time, so that the target candidate is called and no number is assigned.

Since the target classification start threshold value 7 in the classification target determination processing section S 23 has been exceeded in 7 scans, a target number is assigned, and the inter-scan tracking is started from here. But,
Although the threshold value becomes lower than the threshold value 7 in 9 scans, when comparing the feature amounts of 8 scans and 9 scans, since the difference is within a preset range, the two minute targets are regarded as the same and the numbers do not change. Although the threshold value 7 is exceeded in 10 scans, the two minute targets are determined to be the same because the difference is within the preset range when the feature amounts of 9 scans and 10 scans are compared. Since minute targets are continuously extracted, continuous scan-to-scan tracking is possible.

That is, by setting two threshold values, it is possible to reduce the omission of extraction of a target having a large fluctuation in the brightness level. Continuous scan-to-scan tracking can be performed.

It is possible to prevent an increase in the target number for performing classification processing. Processing time is reduced.

[0049]

As described above in detail, according to the present invention, it is possible to reduce the omission of extraction of a target having a large variation in the brightness level, and it is possible to perform continuous scanning-to-scan tracking.

[Brief description of drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is an explanatory view of a main part of the embodiment of the present invention.

FIG. 3 is an operation explanatory diagram of FIG. 2;

FIG. 4 is an explanatory diagram of digital image generation.

FIG. 5 is a main-part configuration diagram of target classification processing means.

FIG. 6 is an explanatory diagram of a target classification method of a conventional example.

7A and 7B are schematic explanatory diagrams of a filter, in which FIG. 7A is an example of a preprocessing filter, and FIG. 7B is an example of an extraction filter.

FIG. 8 is an explanatory diagram of causes of fluctuations in brightness level (in the case of two elements).

[Explanation of symbols]

S 21 Pre-processing filter S 22 Extraction filter S 23 Classification target judgment processing S 24 Target identification section 21 CPU 22 Program RAM 23, 24 Interface 25 RAM

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Hidekazu Nishiuchi 1333 Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture Fujitsu System Integration Laboratories Ltd.

Claims (1)

[Claims] 1. A target categorizing method in an image processing system, which has a CPU and a memory and performs inter-scan tracking for a target satisfying a setting condition among targets extracted from an input digital image. Among the extracted targets having high brightness levels, targets having brightness levels equal to or higher than the first threshold value are extracted, a feature amount including the brightness level of the extracted targets is calculated and stored, and the stored target extraction time Among the brightness levels, it is determined that the target having a brightness level equal to or higher than the second threshold value (second threshold value> first threshold value) is a classification target to be tracked between scans. When the brightness level is lower than the second threshold value, if it is extracted, it is regarded as the same classification target, and the previous and current feature amounts are compared,
A target classification method characterized in that two classification targets are identified as the same target so that they can be tracked if the difference between the feature amounts is within a set range.