KR101429271B1 - Target tracking method and target tracking apparatus using the same - Google Patents

Abstract

The present invention relates to a target tracking method for detecting a target and a target tracking apparatus using the same. The target tracking method comprises the steps of: correcting target images corresponding to a target into the same size, collecting the target images, creating a stochastic target template by performing average operation for the entire collected target images, and creating a target model by producing, as the characteristic information of the target, a ratio of an area, where the brightness of an infrared signal on the created stochastic target template is low, to an area, where the brightness is high; producing, as the characteristic information of an input image, a ratio of an area, where the brightness of an infrared signal on the input image is low, to an area, where the brightness is high, and comparing the produced characteristic information of the input image with the characteristic information of the target to divide into a target candidate group similar to the target and a non-target candidate group not similar to the target; and detecting the target by producing target similarity between the target and the target candidate group using template matching between the target and the target candidate group. Accordingly, the present invention can improve performance in the detection execution speed and removal rate of a non-target candidate group, which is not a target, in an infrared image as compared with existing methods.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a target tracking method and a target tracking method using the same,

The present invention relates to a target tracking method and a target tracking apparatus using the same, and more particularly, to a target tracking method for acquiring and analyzing image information of a moving target in an image, and a target tracking apparatus using the same.

The target tracking device is a device that calculates the trajectory of a target in real time using image information obtained by an image sensor. Object detection in image processing is one of the important topics for tracking the target.

Object detection can be defined as a problem of determining the position of an object to be searched in an image. In order to accurately determine the position of an object, it is important to obtain a feature that can distinguish the characteristic value of the candidate object from the characteristic value of the candidate object.

In other words, it is required to set a feature value that can distinguish between the object to be searched and the non-object. However, since the position of the object to be searched can not be predicted, the detection method must be performed over the entire area.

Since object detection always has a high computational complexity, various methods for reducing computational complexity have been developed. One of these methods is to reduce the number of candidates to be detected by a low-complexity computation before the complexity computation is performed. Through this process, the overall computation speed for object detection can be improved.

However, in spite of this method, there is still insufficient research to improve the overall operation speed for object detection.

KR 10-2010-0041172 A, 2010. 04. 22, Figure 1

It is an object of the present invention to provide a target tracking method and a target tracking apparatus using the target tracking method capable of efficiently removing a non-target candidate group among candidates of a target present in an infrared image to improve the tracking performance of the target.

According to an aspect of the present invention, there is provided a target tracking method comprising: correcting target images corresponding to a target to the same size, collecting the target images, performing averaging on the collected target images, A modeling step of generating a target model by calculating a ratio of a low-brightness region and a high-brightness region of the infrared signal of the generated probabilistic target template as feature information of the target; A ratio of a low-brightness region and a high-region brightness of an infrared signal of an input image is calculated as feature information of the input image, and the feature information of the input image is compared with the feature information of the target, A classification step to classify non-target candidates that are not similar; And a detection step of detecting a target by calculating a target similarity degree between the target and the target candidate group using template matching between the target and the target candidate group.

The feature information may be defined by a patch binary ratio which is a ratio of the number of pixels having a valid brightness value in a predetermined patch divided by the total number of pixels in the patch.

The patch binary ratio can be expressed by Equation (1).

Equation 1:

: 패치 이진 비율,

: 패치의 넓이,

: 패치의 높이,

: 문턱치 값,

: 영상)(here,

: Patch binary ratio,

: The width of the patch,

: Height of patch,

: Threshold value,

: video)

The modeling step may include calculating an adaptive threshold value corresponding to the determined patch binary ratio when the patch binary ratio for the target model is determined and binarizing the probabilistic target template using the calculated threshold value And the feature information can be calculated for the binarized probabilistic target template.

Wherein the classifying step binarizes an input image using an adaptive threshold value calculated by the modeling step, obtains a candidate group patch from the binarized input image, and supplies the patch binary ratio in the obtained candidate group patch to the input image The feature information of the input image and the feature information of the target are compared with each other and classified into the target candidate group and the non-target candidate group.

The template matching may be performed by a probabilistic target template calculated in the modeling step and a target candidate group classified in the classification step. The template matching may be implemented by NCC (Normalized Cross Correlation).

According to another aspect of the present invention, there is provided a target tracking apparatus comprising: a sensor unit for capturing an image of a target and transmitting an input image corresponding to the image of the target; And a signal processing unit for performing the target tracking method according to any one of the first to seventh aspects based on the input image transmitted from the sensor unit.

Accordingly, the present invention rapidly classifies non-target candidate groups and efficiently removes targets and non-target candidate groups before the detection process by using a patch binary ratio with a very high execution speed, Non-target candidate detection performance can be improved in terms of performance and removal rate.

FIG. 1 is a flowchart illustrating a target tracking method according to an embodiment of the present invention. Referring to FIG.
2 is a diagram showing an example of target images used for calculating a target model and a probabilistic target template image obtained using the target images.
FIG. 3 is a diagram showing a change in the target template binarization result according to a change in the threshold value.
FIGS. 4 and 5 are views showing results obtained by removing a candidate group using a conventional method and a method of the present invention in a photographed infrared image.
6 is a schematic block diagram of a target tracking apparatus according to another embodiment of the present invention.

The present invention will now be described with reference to the accompanying drawings. Hereinafter, a repeated description, a known function that may obscure the gist of the present invention, and a detailed description of the configuration will be omitted. Embodiments of the present invention are provided to more fully describe the present invention to those skilled in the art. Thus, the shape and size of the elements in the figures may be exaggerated for clarity.

FIG. 1 is a flowchart illustrating a target tracking method according to an embodiment of the present invention. Referring to FIG.

Referring to FIG. 1, a target tracking method according to an embodiment of the present invention includes a modeling step S10 for creating a model representing a characteristic of an object to be searched, a classification step S20 for classifying a target using a characteristic of a brightness value of infrared rays, , And a detection step S30 for performing detection using template matching for candidate groups related to the object to be searched.

Modeling step (S10) is a step of creating a target model to represent the characteristics of a target, which is an object to be searched. Since the characteristic of the object in the infrared image reflects the signal intensity, that is, the brightness value, it is used as the characteristic value.

In other words, the modeling step (S10) first corrects the images corresponding to the target to the same size, in order to create a target model to represent the characteristics of the target, which is an object to be searched.

The average of all the target images of the same size collected in this way is averaged to generate a single probabilistic target template (model image) (S12). The probabilistic target template (model image) created by performing the averaging operation can be used as a detection template of the detection step S30 described later because it is an image that reflects the brightness characteristic of the target well.

If the patch binary ratio of the target model is determined, the adaptive threshold value corresponding thereto is calculated (S14), and binarization of the probabilistic target template is performed using the threshold value (S15) The characteristic information of the target is calculated (S16).

The feature information of the target can be calculated by the ratio of the low and high brightness regions of the infrared signal of the probabilistic target template binarized by the step S15. That is, the feature information of the target can be determined by the patch binary ratio.

In general, non-point objects in an infrared image have different infrared signal intensities per part (patch), and some of the patches can emit a strong infrared signal.

The patch of the target, which is the object to be searched, has a low region and a high region of the infrared signal, and the ratio of the low region and the high region of the infrared signal may vary depending on the target.

Therefore, it is possible to classify candidates that are not similar to candidates similar to the object to be searched by using the ratio of the low region and the high region as the feature information.

The patch binary ratio represents the characteristic of the infrared brightness value, which is expressed as a ratio of the number of pixels having valid brightness values in the patch divided by the total number of pixels of the patch.

은 패치 이진 비율,

는 패치의 넓이,

는 패치의 높이,

는 적응적 문턱치를 이용해 구한 문턱치,

는 영상을 의미한다.In the above equation

The patch binary ratio,

The width of the patch,

The height of the patch,

Is the threshold value obtained by using the adaptive threshold value,

Means an image.

로 이진화 한 영상의 결과에서 밝은 부분과 전체 영상 크기에 대한 비율로 볼 수 있다. 이와 같이 표적 모델의 패치 이진 비율이 정해지면, 그에 해당하는 적응적 문턱치 값이 계산될 수 있다.That is, the patch binary ratio is the ratio of the image to the adaptive threshold

The results of the binarized image can be seen as a ratio of the brightness to the overall image size. When the patch binary ratio of the target model is thus determined, the corresponding adaptive threshold value can be calculated.

The classification step S20 uses the above-described patch binary ratio to calculate the ratio of the low-brightness region and the high-region brightness of the input image to the feature information of the input image, The feature information of the target described in the modeling step S10 is compared and classified into a target candidate group similar to the target and a non-target candidate group not similar to the target.

The feature information of the input image can be calculated by the patch binary ratio as the feature information of the target mentioned in the modeling step S10.

In the classification step S20, the input image is binarized using the adaptive threshold value calculated in the modeling step S10 (S22), and a candidate patch is obtained from the input image (S24). In the candidate patch patch thus obtained The patch binary ratio, which is the ratio of the number of pixels having the effective brightness value divided by the total number of pixels of the patch, is calculated as feature information of the input image (S26), and the feature information of the input image thus calculated is compared with the feature information of the target Target candidates and non-target candidates (S28).

That is, the input image is binarized (S22) by using the threshold value, and then the patch binary ratio is calculated for all the target candidate groups S24 obtained by the sliding window method (S26), and the patch binary ratio value is 0.3, Is classified as a target candidate group or a non-target candidate group (S28).

The detecting step S30 is a step of detecting the target using the template matching with respect to the target object and the target candidate group to be searched.

The detection step S30 calculates the target similarity by template matching the probabilistic target template calculated at the modeling step S10 and the target candidate group S32 classified at the classification step S20 at step S34, (S36), and non-target or target is detected (S37, S39).

Template matching is a commonly used object detection method. In the experiment process according to the present embodiment, the target similarity between an object to be searched and a target candidate group is determined using NCC (Normalized Cross Correlation) among template matching methods.

For example, template matching determines whether the target and target candidate groups are target or non-target based on the value of the target similarity of 0.5. If the target similarity value is 0.5 or more, it is detected as a target, otherwise as non-target.

Hereinafter, the effect of the target tracking method according to an embodiment of the present invention will be described in comparison with the prior art through experiments. The final executable code for the experiment was implemented in C ++ using OpenCV version 2.4.

In the target tracking method according to the present embodiment, the patch binary ratio used for classification of the non-target candidate group of the infrared image is 0.3. 0.3 means that there is a threshold to distinguish the brightness corresponding to 30% of the object model brightness value to be searched.

The input image is binarized by the corresponding threshold value to compare the patch ratio of the entire candidate patch with the patch binary ratio, and the value corresponding to 0.3 ㅁ 0.25 is classified as the target candidate group. Otherwise, the target candidate group is classified as the non-target candidate group.

The NCC calculation is performed on the remaining target candidates after removal of the non-target candidate group to obtain the similarity. When the similarity value is 0.5 or more, the target is detected. Otherwise, the target is detected as non-target.

Conventional method Invention Number of target candidates before non-target elimination () 37,000 Number of target candidates after non-target removal () 8,614 3,541 Target Removal Ratio (%) 76.7 90.4 Recall (%) 99.0 99.1 Execution time (ms) 1.03 0.83

Table 1 shows the results of comparison between the conventional method and the tracking performance of the present invention. The conventional method shown in Table 1 is a method proposed in Z. Kalal, J. Matas, and K. Mikolajczyk, "P-N learning: Bootstrapping binary classifiers by structural constraints," Computer Vision and Pattern Recognition, It is the result of the method.

The above experimental results are the average of the results of 30 images. The number of total candidates is 37,000, and the conventional method removes about 76.7% (28,386) of non-target candidates among the total candidates (37,000), and the target tracking method according to one embodiment of the present invention is about 90.4% (33,459 Of the non-target candidates were removed.

The recall rate means the ratio of the ground truth to the final detection result, and the higher the recall rate, the more accurate the answer is. It can be seen that both the conventional method and the target tracking method according to the present embodiment require less execution time, and the method according to this embodiment is superior to the conventional method by about 20%.

2 to 5, the effect of the target tracking method according to the present embodiment can be confirmed by comparing with the conventional method.

FIG. 2 is a diagram showing an example of a target image used for calculating a target model and a probabilistic target template image obtained using the target image, FIG. 3 is a diagram showing a result of a target template binarization according to a change in threshold value, 5 is a view showing a result obtained by removing a candidate group from a conventional infrared image and a method of the present invention in a photographed infrared image.

Fig. 2 is an example of images used for calculating a target model and a target template image obtained using the images. 3 is a change in the result of binarization of the target template according to the change in the threshold value.

The leftmost image in FIG. 3 is a template image used for target detection, and the binarization result is shown as the threshold value changes from 0 to 110 from the upper left corner to the lower right corner.

The white area in FIG. 3 is a valid area and the black area is an invalid area. The patch binary ratio is a value obtained by dividing the total number of image pixels by the width of the white region, and has a value between 0 and 1.

FIGS. 4 and 5 are the results of the conventional method and the method of removing the candidate group from the infrared image taken by the method of the present invention.

4 and 5, the red dot is a point at which the candidate group is detected. The spacing of the red dots is set to 1/10 of the horizontal size of the target and 1/10 of the vertical size. One red dot has 10 steps from 15x30, which is 0.5 times the original target size of 30x60, to 42x84, which is 1.4 times the 0.1x interval.

Hereinafter, a target tracking apparatus 1 according to another embodiment of the present invention will be described with reference to FIG. 6 is a block diagram of a target tracking apparatus according to another embodiment of the present invention.

The target tracking apparatus 1 may include a sensor unit 10 that captures an image of a target and transmits an input image corresponding to the image of the captured target, and a signal processing unit 20 that performs the target tracking method described above .

The signal processing unit 20 may include a modeling module 22, a classification module 24, and a detection module 26 for analyzing an input image transmitted from the sensor unit 10 and tracking a target. This distinction is for convenience of description, and can be otherwise divided into different functional modules.

The modeling module 22 corrects the target images corresponding to the target to the same size and collects them, performs averaging operation on the collected target images to generate a probabilistic target template, and generates the probabilistic target template It is possible to perform a function of generating a target model by calculating the ratio of the low-brightness region and the high-region of the infrared signal as the characteristic information of the target.

The classification module 24 calculates the ratio of the low-brightness region and the high-region of the infrared signal of the input image to the feature information of the input image, compares the calculated feature information of the input image with the feature information of the target, It can be classified into similar target candidates and non-target candidates that are not similar to the target.

The detection module 26 can detect the target by calculating the target similarity between the target and the target candidate group using the template matching between the target and the target candidate group.

As described above, the target tracking method and target tracking apparatus 1 according to the present embodiment can quickly classify non-target candidates using a patch binary rate with a very high execution speed, and can efficiently classify target and non- It is possible to improve the performance in the non-target candidate detection speed and the removal rate in the infrared image, which is not a point target.

The target tracking method according to the embodiment of the present invention and the target tracking apparatus 1 using the target tracking method are not limited to the configurations and the methods of the embodiments described above and all or a part of the embodiments may be selectively combined .

The target tracking method and the target tracking apparatus using the target tracking method according to an embodiment of the present invention are not limited to the configurations and methods of the embodiments described above and all or a part of the embodiments may be selectively combined according to the needs of the user .

As described above, the present invention has been described with reference to particular embodiments, such as specific elements, and specific embodiments and drawings. However, it should be understood that the present invention is not limited to the above- And various modifications and changes may be made thereto by those skilled in the art to which the present invention pertains.

It is therefore to be understood that within the scope of the appended claims all such modifications as would be obvious to one skilled in the art are intended to be included within the scope of the appended claims and their equivalents. will be.

1: Target tracking device
10: Sensor unit
20: Signal processor
22: Modeling module
24: Classification module
26: Detection module

Claims (8)

The target image corresponding to the target is corrected and collected to the same size, and an average calculation is performed on all the collected target images to generate a probabilistic target template. Based on the brightness value of the target, the generated probabilistic target A modeling step of generating a target model by calculating a ratio of a low-brightness area and a high-brightness area of the infrared signal of the template as characteristic information of the target;
A ratio of a low-brightness region and a high-region brightness of an infrared signal of an input image is calculated as feature information of the input image, and the feature information of the input image is compared with the feature information of the target, A classification step to classify non-target candidates that are not similar; And
And a detection step of detecting a target by calculating the target similarity degree between the target model and the target candidate group using the target model as a template. The method according to claim 1,
Wherein the characteristic information is determined by a patch binary ratio which is a rate obtained by dividing the number of pixels having a valid brightness value in a predetermined patch by the total number of pixels in the patch. 3. The method of claim 2,
Lt; RTI ID = 0.0 > 1, < / RTI >
Equation 1:

(here,

: Patch binary ratio,

: The width of the patch,

: Height of patch,

: Threshold value,

: video) The method of claim 3,
The modeling step may include calculating an adaptive threshold value corresponding to the determined patch binary ratio when the patch binary ratio for the target model is determined and binarizing the probabilistic target template using the calculated threshold value And the feature information is calculated for the binarized probabilistic target template. 5. The method of claim 4,
Wherein the classifying step binarizes an input image using an adaptive threshold value calculated by the modeling step, obtains a candidate group patch from the binarized input image, and supplies the patch binary ratio in the obtained candidate group patch to the input image Wherein the feature information of the input image is compared with the feature information of the target, and the feature candidate information is classified into the target candidate group and the non-target candidate group. The method according to claim 1,
Wherein the template matching is performed by the probabilistic target template and the target candidate group classified in the classification step using the probabilistic target template calculated in the modeling step as a template. The method according to claim 1,
Wherein the template matching is implemented by NCC (Normalized Cross Correlation). A sensor unit for capturing an image of a target and transmitting an input image corresponding to the image of the captured target; And
And a signal processing unit for performing the target tracking method according to any one of claims 1 to 7 based on the input image transmitted from the sensor unit.

Images