JPH03177983A - Target extraction identification device - Google Patents

Abstract

PURPOSE:To accurately extract a target in spite of luminance difference between the target and a background by obtaining target candidates for plural reference feature quantities by varying a threshold value, respectively, and deciding the target candidate with the minimum distance information out of them as the target. CONSTITUTION:A binarization part 20 binarizes an input image by using a preset threshold value, and a feature quantity calculation part 21 calculates the feature quantity formed based on geometrical shape for a binary image, and a distance calculation part 22 calculates distance between a position in the space of the feature quantity decided with calculated feature quantity and a position in the space of reference feature quantity decided with the dimension item data of plural kinds of targets prepared in advance. A target candidate selection part 23 selects a point having the feature quantity with the minimum value out of calculated distances as the target candidate. A threshold value setting part 24 sets the threshold value variably when the target candidate is selected, and decides the minimum target value by repeating such operation. Thereby, it is possible to surely identify the target even when the difference between the luminance of the target and that of the background is small.

Description

[Detailed Description of the Invention] [Summary] Regarding a target extraction and identification device that is attached to an image acquisition device to extract and identify a certain target on a screen, the difference between the brightness of the target and the brightness of the background is small. A binary part that obtains a binary image using a set threshold, and a feature that calculates feature quantities based on geometric shapes for 21 images, with the aim of reliably identifying the target even in the case of A quantity calculation section;
a distance calculation unit that calculates a distance between a position in a feature space determined by the calculated special @ quantity and a position in a reference feature space determined by dimension specification data of a plurality of types of targets prepared in advance; a target candidate selection unit that selects a point having a special Wi amount that shows the minimum value among the distances selected as a target candidate; a @value variable setting unit that variably sets the threshold value when a target candidate is selected; The target determining unit is configured to determine, as a target, the target candidate having the minimum distance information among the target candidates respectively obtained by the target candidate selecting unit at the threshold value.

[, Industrial application field]

The present invention relates to a target extraction and identification device that is attached to an image acquisition device to extract and identify a certain target on a screen.

2. Description of the Related Art Conventionally, target tracking systems that track a moving target are known, and in this case, the target is extracted and identified from an input image. This input image is used not only when the luminance difference between the target and the background is large, but also in some cases where the luminance difference is small, so the system needs to reliably identify the target regardless of the magnitude of the luminance difference. It is.

[Conventional technology]

FIG. 4 shows a block diagram of a conventional example. In the figure, a threshold value setting section 1 sets a threshold value for an acquired input image based on the respective luminance average values of the background and a target, and a binarization section 2 converts the image into 21 according to the magnitude of the luminance. The binarized input image is subjected to calculation of three types of special quantities, namely the number of pixels (area), aspect ratio, and leader Zhou, in the feature amount calculation unit 3, and these three types of calculated special quantities are sent to the distance calculation unit 4. The distance in the feature space is calculated here. In other words, the reference feature amount (corresponding to the actual size of the target) is obtained by the reference feature suspension part 6 based on the reference specification dimension database 5 in which actual size data of all possible targets is stored.
The distance in the feature amount space (distance on the screen) between the feature amount and the feature amount calculated by the feature amount calculation unit is calculated.

The distance information obtained by the distance calculation section 4 is supplied to the minimum distance detection section 7, where j! I'i distance (on screen) A point having a feature value showing the minimum value is identified as the target.

In this way, in the conventional target extraction and identification device, the threshold value setting unit 1 uniquely determines the intermediate value between the background average brightness level and the target average brightness level as a threshold value, and performs binarization with this same value fixed. Feature values are being calculated.

[Problem to be solved by the invention]

In conventional devices, since the threshold value in the threshold value setting section 1 is fixed, the target brightness level and the background brightness level are in two levels, and the difference between them is large, making it difficult to clearly separate the target and background. If the shape is not in an easy) state, the shape cannot be extracted correctly.
In some cases, there was a problem with the possibility of missing important goals. For example, in infrared images, etc.
If there is a part where the brightness difference between the target (for example, the hood of a Jeep) and the background is small, and a part where the brightness is large between the target (for example, the engine part of a Jeep) and the background where the temperature is high < i & brightness, the brightness A threshold value is set only for a portion with a large difference, and only this portion is binarized, making it impossible to obtain a correct target shape.

SUMMARY OF THE INVENTION An object of the present invention is to provide a target extraction and identification device that can reliably identify a target even when the difference between the brightness of the target and the brightness of the background is small.

[Means to solve the problem]

FIG. 1 shows a diagram of the principle of the present invention. In the figure, 20 is a binarization unit that converts the input image into 2 bits using the set W4IIi.
III to obtain a binary image. 21 is a feature calculation unit;
Calculates feature amounts based on geometric shapes for binary images. A distance calculation unit 22 calculates the distance between ml in the feature space determined by the calculated feature and the position in the reference feature space determined by dimension data of a plurality of types of targets prepared in advance. Reference numeral 23 denotes a target candidate selection unit which selects a point having a feature value representing the minimum value among the calculated distances as an eye m candidate. Reference numeral 24 denotes a threshold i setting unit, which variably sets the threshold value when a target candidate is selected in the target candidate selection unit 23. Reference numeral 25 denotes a target determining unit, which determines as a target the target candidate having the minimum distance information among the target candidates obtained by the target candidate selecting unit at the same value that is variably set.

[For 0] By varying the threshold value, a number of target candidates for a certain reference feature amount are selected, and among them, the one with the smallest distance information is set as the target. Since rAifU is varied to select several target candidates for the complex type of reference feature, compared to the conventional example where the threshold is fixed and once the target candidates are extracted, no other target candidates are extracted. The goal you are trying to find can be extracted more accurately.

〔Example〕

FIG. 2 shows a block diagram of an embodiment of the present invention, in which parts having the same functions as those in FIG. 4 are given the same numbers.

Now, in order to explain and respond to conventional examples, the target will be a jeep. In FIG. 2, the acquired input image is first set with a threshold value based on the luminance average values of the background and the target in a threshold value variable setting section 1o, and then is converted into a binary value according to the magnitude of the luminance in a 2 II conversion section 2. be converted into The binarized input image is subjected to three types of feature quantities ■, ■, ■, which are the number of pixels (area), aspect ratio, and circumference, calculated in the feature quantity calculation unit 3, and the distance 0 output unit 4 calculates three types of feature quantities, as in the conventional example. The distance in the feature space between the point and the reference feature located at is determined, and the minimum distance detection unit 7 identifies the point having the feature indicating the minimum distance value as a target candidate. The obtained target candidates and the threshold values at which they were obtained are stored in the target candidate memory section 11.

When the target candidate and the threshold value at which the target candidate was obtained are stored in the target candidate memory unit 11, a control m+ signal is output from the threshold value control unit 12, thereby indicating that the threshold value set in the threshold value variable setting unit 10 is present. direction. Using this variably set @ value, binarization is performed in exactly the same way as in the previous case, and a feature quantity based on another reference feature quantity is calculated, and the point with the feature quantity showing the minimum distance value is the target point. 11411 and the target candidate at this time are stored in the memory unit 1.
It is stored in 1. In this case, the threshold value is set in advance so that the distance information of each of the sequentially calculated feature amounts becomes smaller.

In this way, when a point with a special @ quantity indicating the minimum distance value with respect to one reference feature quantity is identified as a target candidate,
The threshold value set in the threshold value variable setting unit 10 is varied so that the distance information of the characteristic quantities sequentially calculated by calculating feature quantities based on another reference feature quantity becomes the minimum value. That is, the feature quantities regarding the two-layered image ■, ■,
Feedback is applied to variably set the threshold value for conversion to 21a so that (2) is closest to the reference feature amount for each target category (for example, jeep, truck, etc.).

Figure 3 shows the trajectory of the feature vector when the features ■, ■, ■ are set in a three-dimensional space, and there are two target categories, for example, trucks (target category a) and jeeps (target category b). This is what was written. By varying the threshold value, the special In vector V draws a trajectory (progressing in the direction of the arrow) as shown in the figure, and at a certain threshold value, the closest point A1 to the reference time 'm fl A o of target category a becomes the target candidate. In addition, when a certain threshold value is reached, the closest point B1 to the reference feature amount Bo (the entire shape of the jeep including the engine part and the top part) of the target vehicle b is identified as another target candidate.In this way, The threshold value is varied by applying feedback so that the feature value becomes closest to the reference feature value for each target category, and the target candidate obtained at that time is stored in the target candidate memory section 11.

The target candidates stored in the target candidate memory unit 11 are supplied to the target determining unit 13, which determines the target candidate whose distance information of each target candidate has the minimum value as the final target. In this case, the distance information of (A + Ao) for target category a (truck) and the distance information of (B + Bo) for target category b (jeep) are smaller, so we will use jeep as the target to find target category b. A binary image is obtained, and the optimal R m value at that time is also obtained.In the conventional example, the same value is fixed, so for example, once target category a is extracted as a target candidate, other optimal R m values are also obtained. Although target category b exists, target category b is not extracted, but since the threshold value is variable in the present invention, there is a luminance level difference of three or more levels between the target and the background. Furthermore, even if the difference is small, the target category that is being sought can be extracted accurately.

(Effects of the Invention) As explained above, according to the present invention, target candidates for a plurality of reference feature quantities are obtained by varying the same value, and the target candidate with the minimum distance information is determined as the target from among them.
To accurately extract a desired target even if there is a luminance level difference of three or more levels between the target and the background, or even if the difference is small.

[Brief explanation of drawings]

FIG. 1 is a diagram of the principle of the present invention, FIG. 2 is a block diagram of an embodiment of the present invention, FIG. 3 is a diagram showing the relationship between the feature vector trajectory and the reference feature in the present invention, and FIG. FIG. 1 is a block diagram of a conventional example. In the figure, 2.20 is the binarization section, 3.21 is the feature calculation section, 4.22 is the distance calculation section, 5 is the reference specification dimension database, 6 is the reference feature extraction section, and 7 is the minimum distance. 10.24 is a threshold value variable setting section; 11 is a target candidate memory section; 12t is a tl1m control section; 13.25 is a target determination section; 23 is an eye WA candidate selection section.

Claims (1)

[Claims] A binarization unit (20) that binarizes an input image using a set threshold value to obtain a binary image; A feature amount calculation unit (21) that calculates a feature amount, a position in a feature amount space determined by the calculated feature amount, and a reference feature amount space determined by dimension specification data of a plurality of types of targets prepared in advance. a distance calculation unit (22) that calculates the distance to the position; a target candidate selection unit (23) that selects as a target candidate a point having a feature amount that indicates the minimum value among the calculated distances; a threshold variable setting section (24) that variably sets the threshold when a target candidate is selected in the section (23);
3) A target extraction and identification device comprising a target determining unit (25) for determining, as a target, a target candidate having the minimum distance information among the target candidates obtained in step 3).