KR20180024232A - Method of detecting object in image using selective sliding winsdow and apparatus thereof - Google Patents

Abstract

The present invention relates to a method and an apparatus for detecting an object in an image by using a selective sliding window. The method for detecting an object according to the present invention comprises: a step of receiving a taken image from a camera; a step of sliding a unit window as much as integer times of the number of movement pixels preset in the taken image, to calculate first classification result values for a plurality of first image areas corresponding to the slid unit window; a step of selecting a first image area where the first classification result value is larger than a reference object detection probability value; a step of selecting an image area corresponding to the unit window slid as much as a preset number of movement pixels in at least one direction of up, down, left, and right from the selected first area, as a second image area; a step of calculating a second classification result value for the plurality of second image areas; and a step of detecting an object in the taken image by using the first and second classification result values. According to the present invention, since calculation is performed on an object search area for object detection selected from the whole image area, it is possible to reduce the time required to detect an object while maintaining the same detection performance.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a method for detecting an object in an image using a selective sliding window,

The present invention relates to a method and an apparatus for detecting an object in an image using a selective sliding window, and more particularly, to a method and an apparatus for detecting an object in an image using a selective sliding window will be.

Recently, a method of detecting a specific object such as a person, a vehicle, a two-wheeled vehicle, etc. using an image in the fields of intelligent automobile, robot, and image security has been actively researched. Object detection technology refers to a technique for estimating the position and size of a specific object such as a person or a vehicle in a given image.

1 is a view for explaining a conventional object detection technique. As shown in FIG. 1, a general object detection technique searches for an entire region of an image using a classifier learned in advance, and determines the presence or absence of the object. Since the model learned in the classifier generally has a window size of a predetermined size, in order to detect objects of various sizes in the image, it is necessary to search while changing the size of the test image. In order to estimate the position of the object, You must navigate the object while moving the window.

In order to precisely detect an object using this multi-scale sliding window method, it is necessary to move the area to be searched finely. However, this increases the number of windows to be searched, .

Accordingly, in order to reduce the computation time by minimizing the search area, methods of extracting the region of interest by distinguishing the background region and the foreground region such as the background model, the motion model, and the 3D modeling technique have been proposed.

However, object detection methods based on ROIs require preliminary information related to camera information or images, and there are many restrictions on application fields because ROIs can be extracted only under limited conditions. In addition, if an undetected or incorrect region of interest occurs due to an error in the region of interest estimation, a problem that an object can not be detected may occur.

Therefore, there is a need for an object detection technique capable of not only precisely extracting an object from an image but also improving the operation speed without any restriction.

The background technology of the present invention is disclosed in Korean Patent Publication No. 10-2016-0046210 (published on June 28, 2016).

An object of the present invention is to provide a method and an apparatus for detecting an object in an image using a selective sliding window for improving an operation speed when detecting an object in an image.

According to another aspect of the present invention, there is provided a method of detecting an object in an image using an object detecting apparatus, the method comprising: receiving a captured image from a camera; Calculating a first classification result value for a plurality of first image regions corresponding to the sliding unit window by sliding the unit window as many times as the first window, Selecting an image region corresponding to a unit window that has been slid by a predetermined number of movement pixels in at least one direction of the upper side, the lower side, the left side, and the right side from the selected first image region as a second image region; , Calculating second classification result values for the plurality of second video regions, And detecting an object in the photographed image using the first and second classification result values.

The integer multiple may be an integral multiple of 2 or more.

Wherein the step of calculating the first classification result values comprises the steps of: sliding a unit window by an integral multiple of the predetermined number of moving pixels in the captured image, calculating a feature vector of a plurality of first image areas corresponding to the unit window Extracting feature vectors corresponding to the plurality of first image regions, and calculating first classification result values corresponding to the plurality of first image regions by inputting feature vectors for the plurality of first image regions to the previously learned classifiers.

The calculating of the second classification result values may include extracting the feature vectors for the plurality of second image regions, inputting the feature vectors for the plurality of second image regions to the learned classifier, And calculate second classification result values corresponding to the plurality of second video regions.

The method may further include converting the input captured image by a predetermined image size to detect objects of various sizes.

According to another aspect of the present invention, there is provided an apparatus for detecting an object, the apparatus comprising: an input unit for inputting a photographed image from a camera; a unit window for sliding an integral number of the predetermined number of moving pixels in the captured image, A first calculation unit for calculating first classification results for a first image region and a second classification result for a first image region, and a second calculation unit for selecting a first image region having a first classification result value larger than a reference object detection probability value, A selection unit for selecting, as a second image region, an image region corresponding to a unit window that is slid by a predetermined number of movement pixels in at least one direction of a left side and a right side, And a detection unit for detecting an object in the photographed image using the first and second classification result values, .

As described above, according to the present invention, since the object search area for object detection is selected and operated in the entire image area, the time required for object detection can be reduced while maintaining the same detection performance.

1 is a view for explaining a conventional object detection technique.
2 is a configuration diagram of an object detecting apparatus according to an embodiment of the present invention.
3 is a flowchart of an object detection method according to an embodiment of the present invention.
FIG. 4 is a diagram for explaining step S320 of FIG. 3 in detail.
FIG. 5 is a diagram for explaining step S340 of FIG. 3 in detail.
6 is a view for explaining an image search area selection process according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily carry out the present invention. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. In order to clearly illustrate the present invention, parts not related to the description are omitted, and similar parts are denoted by like reference characters throughout the specification.

Throughout the specification, when an element is referred to as "comprising ", it means that it can include other elements as well, without excluding other elements unless specifically stated otherwise.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily carry out the present invention.

First, an object detection apparatus using a selective sliding window according to an embodiment of the present invention will be described with reference to FIG. 2 is a configuration diagram of an object detecting apparatus according to an embodiment of the present invention.

2, the object detecting apparatus 100 according to the embodiment of the present invention includes an input unit 110, a first calculating unit 120, a selecting unit 130, a second calculating unit 140, 150, and may further include a conversion unit 160.

First, the input unit 110 receives a photographed image from a camera.

Next, the first calculation unit 120 calculates a first classification result value for a plurality of first image regions corresponding to the sliding unit window by sliding a unit window by an integral multiple of the predetermined number of moving pixels in the captured image do.

Specifically, the first calculation unit 120 slides the unit window by an integral multiple of the predetermined number of moving pixels in the captured image. The first calculation unit 120 extracts feature vectors for a plurality of first image regions corresponding to the unit window that has been slid by an integral multiple. In this case, an integer multiple means an integer multiple of 2 or more.

Then, the first calculation unit 120 calculates the first classification result values corresponding to the plurality of first image regions by inputting the feature vectors for the plurality of first image regions to the learned classifier.

Next, the selecting unit 130 selects the first image area having the first classification result value larger than the reference object detection probability value.

Then, the selection unit 130 selects, as the second video region, the video region corresponding to the unit window that has been slid by the predetermined number of the moving pixels in at least one of the upper side, the lower side, the left side, and the right side from the selected first video region .

Next, the second calculation unit 140 calculates second classification result values for the plurality of second video regions.

Specifically, the second calculator 140 extracts feature vectors for a plurality of second image regions. The second calculation unit 140 calculates the second classification result values corresponding to the plurality of second image regions by inputting the feature vectors of the plurality of second image regions to the learned classifier.

Meanwhile, the classifier learned in the first calculation unit 120 or the second calculation unit 140 means an object detection classifier capable of determining whether there is an object to be detected in the corresponding image region. The classifier includes a SVM classifier (Support Vector Machine Classifier), and may further include a classifier for detecting other objects.

Next, the detection unit 150 detects an object in the photographed image using the first and second classification result values.

Next, the conversion unit 160 converts the input captured image by predetermined image size to detect objects of various sizes.

Hereinafter, a method of detecting an object in an image using a selective sliding window according to an exemplary embodiment of the present invention will be described with reference to FIG. 3 through FIG. 3 is a flowchart of an object detection method according to an embodiment of the present invention.

First, the input unit 110 receives the photographed image from the camera (S310).

Next, the first calculation unit 120 calculates a first classification result value for a plurality of first image regions corresponding to the sliding unit window by sliding a unit window by an integral multiple of the predetermined number of moving pixels in the captured image (S320).

FIG. 4 is a view for explaining step S320 of FIG. 3 in detail. FIG. 4 (a) illustrates a process of sliding a unit window, and FIG. 4 (b) 1 < / RTI >

Specifically, the first calculation unit 120 slides the unit window by an integral multiple of the predetermined number of moving pixels in the captured image. In this case, an integer multiple means an integer multiple of 2 or more.

For example, it is assumed that the unit window is slid by twice the predetermined number of moving pixels. Then, the first calculation unit 120 moves the unit window to the right and to the lower side by twice the predetermined number of moving pixels starting from the upper left end of the image as shown in FIG. 4 (a). Here, the grid displayed on the image of FIG. 4 indicates the predetermined number of moving pixels.

The first calculation unit 120 extracts feature vectors for a plurality of first image regions corresponding to the unit window that has been slid by an integral multiple. For example, when the unit window is slid as shown in FIG. 4A, the first calculation unit 120 extracts 36 feature vectors for the first image region, as shown in FIG. 4B. do.

Here, the feature vector refers to a parameter representing an image feature of the image region, and includes a feature vector such as HOG (Histogram Of Gradients Feature).

Then, the first calculation unit 120 calculates the first classification result values corresponding to the plurality of first image regions by inputting the feature vectors for the plurality of first image regions to the learned classifier.

For example, the first calculation unit 120 may calculate a first classification result value of the first image region by inputting a feature vector such as HOG of the first image region to a classifier such as SVM.

Then, the selecting unit 130 selects a first image area having a first classification result value larger than the reference object detection probability value (S330). At this time, the reference object detection probability value can be set differently according to the object to be detected in the image, and the design can be changed by a person skilled in the art.

The selecting unit 130 may select one of the plurality of video regions corresponding to the unit window that is slid by the predetermined number of moving pixels in the direction of at least one of the upper side, the lower side, the left side, and the right side from the first video region selected in operation S330. And selects image areas (S340). FIG. 5 is a diagram for explaining step S340 of FIG. 3 in detail.

For example, the box marked O in FIG. 5 represents the left-top of the first image area selected in step S330, i.e., the first image area where the first classification result value is larger than the reference object detection probability value. At this time, as shown in FIG. 5, the selecting unit 130 selects the video region corresponding to the unit window that is slid by the predetermined number of the moving pixels to the upper side, the lower side, the left side, and the right side of the selected first video region as the second video region do. In FIG. 5, nine second video regions are selected.

Next, the second calculation unit 140 calculates second classification result values for the plurality of second video regions (S350).

Specifically, the second calculator 140 extracts feature vectors for a plurality of second image regions. The second calculation unit 140 calculates the second classification result values corresponding to the plurality of second image regions by inputting the feature vectors of the plurality of second image regions to the learned classifier.

Then, the detecting unit 150 detects the object in the photographed image using the first and second classification result values (S360). For example, the detection unit 150 may detect an object in the photographed image by removing a low-confidence classification result value among the first and second classification result values using a non-maximum suppression (NMS) technique.

On the other hand, the converting unit 160 may convert the input captured image according to a preset image size (not shown). Specifically, the converting unit 160 may convert the captured image input after step S310 according to a preset image size. Then, the object detecting apparatus 100 according to the embodiment of the present invention performs steps S320 to S360 for each shot image converted for each image size to detect objects of various sizes.

On the other hand, the object detecting apparatus 100 according to the embodiment of the present invention can further select an image search area for object detection using a second classification result value (not shown). 6 is a view for explaining an image search area selection process according to an embodiment of the present invention.

As shown in FIG. 6, the object detecting apparatus 100 according to the embodiment of the present invention can select an area having a classification result value of the second image area that is larger than the reference object detection probability value.

Then, the object detecting apparatus 100 detects an image region corresponding to a unit window that is slid by the predetermined number of movement pixels from the second image region whose classification result value is larger than the reference object detection probability value to the upper, lower, left, 3 Select the image area. Then, the object detecting apparatus 100 calculates the classification result value of the third video region.

Next, the object detecting apparatus 100 repeatedly performs the above-described image search region selection process until there is no selectable region as the classification result value of the third image region.

Then, as shown in FIG. 6, the object detecting apparatus 100 detects 54 objects in the image by searching only 54 image areas out of 126 image areas through the above process.

According to the embodiment of the present invention, since the object search area for object detection is selected and operated in the entire image area, the time required for object detection can be reduced while maintaining the same detection performance.

While the present invention has been described with reference to exemplary embodiments, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

100: object detecting apparatus 110: inputting unit
120: first calculation unit 130:
140: second calculating section 150: detecting section
160:

Claims (10)

A method for detecting an object in an image using an object detecting apparatus,
Receiving a photographed image from a camera,
Calculating first classification result values for a plurality of first image regions corresponding to the sliding unit window by sliding a unit window by integer times the predetermined number of moving pixels in the captured image,
Selecting a first image region in which the first classification result value is larger than a reference object detection probability value,
Selecting an image area corresponding to a unit window that is slid by a predetermined number of movement pixels in at least one direction of the upper side, the lower side, the left side, and the right side from the selected first image area as a second image area,
Calculating second classification result values for the plurality of second video regions, and
And detecting an object in the photographed image using the first and second classification result values. The method according to claim 1,
Wherein the integer multiple is an integer multiple of 2 or more. The method according to claim 1,
The step of calculating the first classification result values comprises:
Extracting feature vectors for a plurality of first image regions corresponding to a unit window that has been slid by the integral multiple by sliding a unit window by an integral multiple of the predetermined number of moving pixels in the captured image,
And inputting the feature vectors for the plurality of first image regions into the learned classifier to calculate first classification result values corresponding to the plurality of first image regions. The method according to claim 1,
Wherein the step of calculating the second classification result values comprises:
Extracting feature vectors for the plurality of second video regions, and
And inputting a feature vector for the plurality of second image regions to the learned classifier to calculate second classification result values corresponding to the plurality of second image regions. The method according to claim 1,
Further comprising the step of transforming the input captured image by a predetermined image size to detect objects of various sizes. An input unit for inputting the photographed image from the camera,
A first calculation unit for calculating a first classification result value for a plurality of first image regions corresponding to the sliding unit window by sliding a unit window by an integral multiple of the predetermined number of moving pixels in the captured image,
A first image region having a first classification result value larger than a reference object detection probability value and a second image region having a first classification result value larger than a reference object detection probability value, A selection unit for selecting a video area corresponding to a unit window as a second video area,
A second calculation unit for calculating second classification result values for the plurality of second image areas, and
And a detector for detecting an object in the photographed image using the first and second classification result values. The method according to claim 6,
Wherein the integer multiple is an integer multiple of 2 or more. The method according to claim 6,
The first calculation unit calculates,
A unit window is slid by an integral multiple of the predetermined number of moving pixels in the captured image to extract feature vectors for a plurality of first image regions corresponding to the unit window that has been slid by integer times, And inputting the feature vectors for the first image region to calculate first classification result values corresponding to the plurality of first image regions. The method according to claim 6,
The second calculation unit calculates,
Extracting feature vectors for the plurality of second image regions, inputting a feature vector for the plurality of second image regions to the learned classifier, and outputting a second classification corresponding to the plurality of second image regions, And calculates result values. The method according to claim 6,
Further comprising: a conversion unit for converting the input captured image according to a predetermined image size to detect objects of various sizes.

Images