KR101313879B1 - Detecting and Tracing System of Human Using Gradient Histogram and Method of The Same - Google Patents

Abstract

A motion region detector for detecting a motion region, which is a candidate region for human detection, from a captured image, a human region recognition unit for recognizing whether or not the movement region is a human region, and a recognized person so that robust tracking can be performed even in the overlapping phenomenon between people. A human area tracking unit which tracks an area, wherein the human area recognition unit includes a gradient image generation unit that generates a pixel-level gradient image, a gradient calculation unit that calculates a direction and magnitude of the pixel gradient, and normalizes gradient information in cell units. Provided is a human detection tracking system using a gradient histogram including a gradient information normalization unit and a feature vector generator for obtaining a histogram of each cell for each slope and deriving feature vectors of the obtained histogram values.

Description

{Detecting and Tracing System of Human Using Gradient Histogram and Method of The Same}

The present invention relates to a human detection tracking system and method using a tilt histogram, and more particularly, to segment a moving object region from video data input from a video acquired under a fixed camera environment, and to detect a person in the moving object region. A human detection tracking system using a tilt histogram that generates a plurality of tilt images having different sizes from head to shoulder, an area, obtains a support vector of tilt directional feature vectors, and detects a person using learning of a support vector machine; It is about a method.

In general, a method of detecting a person is to detect a person by setting a candidate region of the person detection as a whole body region of the person and extracting a feature vector, or by setting a candidate region as a head and extracting a feature vector. Method and the like.

In the method of setting a candidate region for detecting a person as a whole body region of the person, a method of detecting a person by generating a gradient histogram feature vector, which is excellent in detection performance because of using the whole body region as a feature There is a limitation in solving the phenomenon, and in the case of a high resolution image, it takes a lot of processing time to extract the feature vector for human detection.

In the case of setting the candidate region as the head part, the process of extracting and detecting the feature vector is excellent, but there is a problem that the false detection rate is high because the feature for detecting a person is not clear.

The present invention has been made in view of the above points, and generates a plurality of tilted images having different sizes from head to shoulder, which are candidate areas for human detection, in a fixed camera environment so that robust tracking can be performed even in the overlapping phenomenon between people. Then, to provide a human detection tracking system and method using a tilt histogram that can detect and track a human region based on the directional feature vectors of the slope, an object thereof.

The human detection tracking method using a tilt histogram according to an embodiment of the present invention divides a region of a moving object from video data input from a video acquired under a fixed camera environment, and uses a head as a candidate region of a human detection to a moving object region. Generate a plurality of tilt images of different sizes to the shoulder, obtain a support vector using the directional feature vectors of the tilt image, and learn a person by using a support vector machine. Detection process.

In addition, the human detection tracking method using the tilt histogram according to an embodiment of the present invention may further include a step of tracking only a person in a continuous image using the center coordinates of the detected human region.

In addition, the human detection tracking system using the tilt histogram according to an embodiment of the present invention is a motion region detection unit for detecting a motion region that is a candidate region for human detection from the captured image, and a human region for recognizing whether the motion region is a human region or not. And a person zone tracking unit for tracking the recognized person zone.

The human region recognition unit includes a tilt image generator for generating a pixel-based tilt image, a tilt calculator for calculating a direction and magnitude of pixel tilt, a tilt information normalizer for normalizing tilt information of a cell unit, and each tilt And a feature vector generator for obtaining histograms of cells and deriving feature vectors of the obtained histogram values.

The human region recognition unit recognizes whether the motion region is a person or not based on the feature vectors of the tilted images.

The tilt image generating unit extracts an image of an area from the image of the motion detection unit to the head to shoulder which is the human detection candidate region into a plurality of images of the same region, and then generates each extracted image as a tilt image.

The human region recognition unit compares a value of a feature vector, which is a recognition criterion of the human region, with a value of support vectors, from video or video data acquired in an arbitrary fixed camera environment, and compares one of the feature vectors. When the feature vectors of U are similar to each other by a certain probability value or more, the motion region is recognized as a person.

According to the human detection tracking system and method using the tilt histogram according to an embodiment of the present invention, in a fixed camera environment, using a tilt histogram of the head-to-shoulder area, which is a human detection candidate area, detects and tracks a person efficiently, Robust tracking can be performed even in the overlapping phenomenon.

In addition, even in the occlusion phenomenon between people, it is possible to greatly improve the recognition rate of the human domain by recognizing and tracking people.

1 is a block diagram schematically illustrating a human detection tracking system using a slope histogram according to an embodiment of the present invention.
2 is a block diagram schematically illustrating a configuration of a human region recognition unit in a human detection tracking system using a tilt histogram according to an embodiment of the present invention.
3 is a block diagram illustrating in detail a configuration of a human area tracking unit in a human detection tracking system using a tilt histogram according to an embodiment of the present invention.
4 is a flowchart schematically illustrating a person detection tracking method using a slope histogram according to another embodiment of the present invention.

Next, a preferred embodiment of a human detection tracking system and method using a tilt histogram according to the present invention will be described in detail with reference to the accompanying drawings. The present invention can be embodied in various forms and is not limited to the embodiments described below.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the accompanying drawings, wherein like numerals refer to like elements throughout.

First, as shown in FIG. 1, a person detection tracking system using a tilt histogram according to an embodiment of the present invention includes a video acquisition unit 10, a motion area detection unit 20, a human area recognition unit 30, , The human area tracking unit 40 is included.

Since the video acquiring unit 10 may be generally implemented by applying a closed circuit TV (CC-TV) or a CCD camera used to capture a video image, detailed description thereof will be omitted.

The video acquisition unit 10 acquires a video in a fixed environment of the camera and transmits the video to the motion region detection unit 20. That is, the video acquisition unit 10 generates a video signal and transmits the video signal to the motion region detection unit 20.

The motion area detector 20 detects a motion area that is a candidate area for human detection from an image photographed by the video acquisition unit 10.

The movement area detector 20 divides the area of the moving object and transfers the area of the moving object to the human area recognizer 30.

The motion region detector 20 calculates statistical characteristics of the background region from screens of a certain length having only a background, and then compares the statistical characteristics of the newly input image with each other to detect pixels determined to have motion. The pixels are divided into pixels by using a connected component labeling technique, and the moving region is detected for each screen to detect the moving region.

The human area recognition unit 30 is configured to recognize whether or not the movement area is a human area.

The human region recognition unit 30 generates a slope histogram of a region from the head to the shoulder, which is a candidate region for human detection, obtains the direction and magnitude of the pixel-by-pixel gradient, and calculates the extracted value of the direction and magnitude of the pixel-by-pixel gradient. The feature vector is obtained through the process of normalization by unit, and compared with a predefined human domain feature vector, it is recognized as a person when it is similar by more than a certain probability value.

As shown in FIG. 2, the human region recognition unit 30 includes a gradient image generation unit 32 for generating a pixel-by-pixel gradient image, a gradient calculation unit 34 for calculating the direction and magnitude of the pixel slope, and a cell. It may also include a gradient information normalizer 36 for normalizing gradient information of a unit and a feature vector generator 38 for obtaining a histogram of each cell for each slope and deriving feature vectors of the obtained histogram values.

The human region recognition unit 30 recognizes whether the movement region is a person or not based on the feature vectors of the tilted images.

The tilt image generator 32 extracts an image of a region from the image of the movement region detector 20 to the head to shoulder, which is the human detection candidate region, into a plurality of images of the same region, and then extracts each extracted image. Generate as tilt image.

The human region recognizing unit 30 determines the value of a feature vector and a support vector (predefined human region feature vector) that are recognition criteria of the human region from video or video data acquired in an arbitrary fixed camera environment. Are compared with each other, and the motion region is recognized as a person when one feature vector of the feature vectors is similar to a certain human image by a predetermined probability value or more.

The person region tracking unit 40 tracks the recognized person region by calculating the center coordinates of the person region transmitted from the person region recognition unit 30 as a trajectory of the person region over time.

As shown in FIG. 3, the human region tracking unit 40 includes a tracking instruction unit 41, a motion model parameter unit 42, a motion prediction unit 43, a blinding inference unit 44, and a position. It comprises a size measuring section 45, and the motion parameter update unit 46.

The tracking instructing unit 41 transmits the center coordinates and the movement speed of the human area to the motion model parameter unit 42 when a command for starting the tracking is given.

The motion model parameter unit 42 stores and parameterizes each person by using similarity between the images of the center coordinates and the movement speed from the newly input images.

The motion model parameter unit 42 transmits each parameter to the motion predictor 43 with acquisition of a new image.

The motion estimator 43 transmits parameter values to the occlusion inference unit 44 to predict coordinates to be moved in a new image for each person and to predict an occlusion phenomenon.

When the occlusion phenomenon occurs, the occlusion phenomenon inference unit 44 predicts the coordinate value after the occlusion phenomenon in a new image and transmits the coordinate value to the position size measuring unit 45.

The position size measuring unit 45 measures a position and a size to determine whether a person who appears in the new image is the same person as the person in the previous image by comparing with a parameter value before the new image.

The value measured by the position size measuring unit 45 is transmitted to the motion parameter updating unit 46 to update the parameter value.

The human area tracking unit 40 may further include a motion pattern analyzer 47 and a motion descriptor generator 48.

The motion pattern analyzer 47 analyzes the trajectory and the movement speed of a person using the measured position and size values transmitted from the position size measuring unit 45 and transmits the trajectory and the moving speed to the motion descriptor generator 48. .

The motion descriptor generator 48 indexes the motion trajectory and the moving speed so as to be searchable even after the tracking of each person, and returns data.

According to another embodiment of the present invention, a method for tracking human detection using a tilt histogram, as illustrated in FIG. 4, may be performed by dividing a region (S10), generating a tilt image (S20), and obtaining a support vector. Step S30, and detecting the person (S40).

In the step of dividing the area (S10), the area of the moving object is divided from the video data input from the video acquired under the fixed camera environment.

In the step of generating the tilt image (S20), a plurality of tilt images having different sizes are generated from the head to the shoulder, which are candidate regions for human detection of the moving object region.

In the generating of the tilted image (S20), after extracting the image of the region from the motion detection image divided in the step S10 of the head to the shoulder, which is a human detection candidate region, into a plurality of images of the same region Each extracted image is generated as a gradient image.

In operation S30, the support vector is obtained using the directional feature vectors of the tilted image.

The support vector obtained above is used as a predefined human region feature vector.

In the detecting of the person (S40), the person is detected using learning of a support vector machine.

For example, in the step of generating the tilt image (S20), a tilt histogram of a region from the head to the shoulder, which is a candidate region for human detection, is generated to obtain a direction and magnitude of the pixel-by-pixel tilt, and to obtain the support vector (S30). A feature vector is obtained through a process of normalizing the direction and magnitude of the pixel unit gradient extracted in units of cells, and is defined as a support vector.

When the support vector defined as described above and the feature vector of the newly input video image are compared and similar to each other by a predetermined probability value or more, the person is recognized in the step S40 of detecting the person.

In the detecting of the person (S40), a value of a feature vector and a support vector (a predefined person region feature vector), which are recognition criteria of a person region acquired from video or video data acquired in an arbitrary fixed camera environment, are defined. By comparing these values with each other, when a feature vector of one of the feature vectors is similar to a certain human image by a predetermined probability value or more, the motion region is recognized as a person.

In addition, the human detection tracking method using the tilt histogram according to another embodiment of the present invention may further include the step S50 of tracking only a person in a continuous image using the center coordinates of the detected human region.

In the tracking of the person (S50), the center coordinates of the human area are calculated as the trajectory of the human area over time to track the recognized human area.

For example, in the step of tracking the person (S50), when a command for the start of the tracking is given, the similarity between the center coordinates and the movement speed of the human area is similar to each other using the similarity between the existing images from the newly input images. Parameterize, predict each person's coordinates to move in the new image, and if the occlusion occurs, predict the coordinate value after the occlusion in the new image and compare it with the parameter value before the new image. In order to determine whether the person displayed is the same person as the person in the previous image, the position and size are measured, the parameter values are updated with the measured values, and the trajectory and movement of the person are respectively measured using the measured position and size values. Analyze the speed and index the movement trajectory and movement speed so that you can search after each person's tracking. Returns data

In the above, a preferred embodiment of a human detection tracking system and method using a tilt histogram according to the present invention has been described, but the present invention is not limited thereto, and various modifications are made within the scope of the claims, the specification, and the accompanying drawings. It is possible to do this and this also belongs to the scope of the present invention.

10-video acquisition unit, 20-motion area detection unit, 30-human area recognition unit
32-tilt image generator, 34-tilt calculator, 36-tilt information normalizer
38-feature vector generator, 40-human area tracking, 41-tracking indicator
42-motion model parameter section, 43-motion prediction section, 44-occlusion inference section
45-Position size measurement section, 46-Motion parameter update section
47-motion pattern analyzer, 48-motion descriptor generator

Claims (10)

A motion region detector for detecting a motion region that is a candidate region for human detection from the captured image, a human region recognizer for recognizing whether the motion region is a human region, and a human region tracker for tracking the recognized human region,
The human region recognition unit includes a tilt image generator for generating a pixel-based tilt image, a tilt calculator for calculating a direction and magnitude of pixel tilt, a tilt information normalizer for normalizing tilt information of a cell unit, and each tilt A feature vector generator for obtaining a histogram of cells and deriving feature vectors of the obtained histogram values;
The human region recognition unit recognizes whether or not the movement region is a person based on the feature vector of the tilted images.
The motion area detection unit calculates statistical properties of the background area from screens of a certain length having only a background, and then compares the statistical properties of the newly input image with each other to detect pixels that are judged to have motion. A human detection tracking system using a tilt histogram for dividing an area of a moving object using a connected component labeling technique and detecting the moving area by performing such an operation every screen. The method according to claim 1,
The tilt image generating unit extracts an image of the region from the head to the shoulder, which is the human detection candidate region, into a plurality of images of the same region from the image of the movement region detecting unit, and then generates each of the extracted images as a tilt image. Human detection tracking system using histogram. The method according to claim 1,
The human region recognition unit compares the value of the feature vector, which is a recognition criterion of the human region, with the values of the support vectors, from video or video data acquired in an arbitrary fixed camera environment, so that one feature vector of the feature vectors is obtained. A human detection tracking system using a slope histogram that recognizes the movement area as a person when the image is similar to a predetermined probability value of a certain human image. delete The method according to claim 1,
The human region tracking unit includes a tracking instruction unit, a motion model parameter unit, a motion prediction unit, a blinding inference unit, a position size measurement unit, and a motion parameter update unit.
The tracking command unit transmits the center coordinates and the movement speed of the human area to the motion model parameter unit when a command for starting the tracking is given.
The motion model parameter unit stores and parameterizes each person from the newly input images by using similarity between the images of the center coordinates and the movement speed, and transmits each parameter to the motion prediction unit along with the acquisition of a new image. and,
The motion prediction unit predicts coordinates to be moved in a new image for each person and transmits parameter values to the occlusion inference unit to predict an occlusion phenomenon.
When the occlusion phenomenon occurs, the occlusion inference unit predicts the coordinate value after the occlusion in a new image and transmits the coordinate value to the position size measurement unit.
The position size measuring unit measures the position and size to determine whether a person who is represented in the new image is the same person as the person in the previous image by comparing with a parameter value of the previous image.
And the motion parameter updater uses a slope histogram to update the parameter value with the value measured by the position size measurer. The method according to claim 5,
The human region tracking unit includes a motion pattern analyzer for analyzing a person's trajectory and a moving speed by using the measured position and size values transmitted from the position size measuring unit, and a motion trajectory and a movement transmitted from the motion pattern analyzer. A person detection tracking system using a slope histogram further comprising a motion descriptor generation unit for indexing the speed so that a search is possible even after the tracking of each person and returning data. Divides the area of the moving object from the video data input from the video acquired under the fixed camera environment,
Generates a plurality of gradient images of different sizes from head to shoulder, which are candidate areas for human detection of a moving object region,
A support vector is obtained using the directional feature vectors of the tilted image.
A process of detecting a person using a learning of a support vector machine and a process of tracking only a person in a continuous image using the center coordinates of the detected person region,
In the process of tracking a person, when a command to start the tracking is given, the similarity between the center coordinates of the human area and the movement speed is parameterized for each person by using the similarity between the existing images from the newly inputted images. Predict the coordinates to be moved in the new image every time, and if the occlusion occurs, predict the coordinate value after the occlusion in the new image, and compare the parameter value with the previous parameter to the new image. Measure the position and size to determine whether the person is the same person, update the parameter value with the measured value, analyze the trajectory and moving speed of each person using the measured position and size values, respectively. Return the data by indexing the movement trajectory and movement speed so that the search can be performed even after tracking Human detection tracking method using the slope histogram comprising the step of. The method of claim 7,
When generating the tilted image, an image of the region from the divided moving object to the head-to-shoulder, which is a human detection candidate region, is extracted as a plurality of images of the same region, and then each tilted image is generated as a tilted image. Human detection tracking method using histogram. The method of claim 7,
By comparing the value of the feature vector, which is a recognition criterion of the human region acquired from the video or video data acquired in an arbitrary fixed camera environment, and the value of the support vectors, one feature vector of the feature vectors is compared with an arbitrary person image. A human detection tracking method using a slope histogram that recognizes a moving area as a person when it is similar to a predetermined probability value or more. delete

Images