KR100566629B1 - System for detecting moving objects and method thereof - Google Patents

Abstract

The moving object detection system and method of the present invention integrates a cumulative difference image and a motion estimation technique to automatically detect a moving object in a continuous image obtained by a moving camera, thereby solving the problem of initiating a conventional moving object tracking system at low cost. To provide a moving object detection system and method for completing the automation of the tracking system.

In order to achieve the above object, the present invention provides a motion information acquisition unit that sequentially receives a sequence image frame, obtains a cumulative difference image between frames, and generates a motion image data by performing a morphological filtering process with the cumulative difference image. ; And a final candidate block detector which receives the sequence image frame and the motion image data, estimates motion through a block matching process, and detects a final candidate block area according to the estimated result.

Cumulative difference video, motion estimation, video

Description

Moving object detection system and method {SYSTEM FOR DETECTING MOVING OBJECTS AND METHOD THEREOF}             

1 is a flow chart showing a conventional moving object detection method,

2 is a block diagram showing a conventional moving object detection system;

3 is a block diagram showing a moving object detection system according to an embodiment of the present invention;

4A is an exemplary diagram illustrating a cumulative difference image using three initial frames;

4B is an exemplary diagram illustrating an image in which motion information is accumulated for nine consecutive frames;

4c is an exemplary view showing an image from which noise is removed by using a morphological filter;

5 is an exemplary view showing a block matching method of a moving object detection system according to an embodiment of the present invention;

6A is an exemplary diagram illustrating a motion information image and block division;

6B is an exemplary diagram illustrating a block initialized image;

6C is an exemplary diagram illustrating candidate blocks after three frames;

6D is an exemplary diagram illustrating final candidate blocks after nine frames;

7A and 7B are exemplary views illustrating detection of a moving object using XY-projection in a search window area;

8 is a flowchart illustrating a moving object detection method according to an embodiment of the present invention.

Explanation of symbols on the main parts of the drawings

310: motion information acquisition unit 330: final candidate block detection unit

The present invention relates to a moving object detection system and method. In particular, the candidate region obtained by integrating a cumulative difference image and a motion estimation technique is estimated in a block unit in a continuous image, and a portion where the estimated change of motion is severely determined. A moving object detection system and method for detecting a moving object while removing it.

Recently, research on computer vision has attempted to build a system that replaces human visual ability in the natural environment. Until now, it has been progressed in the laboratory environment, which is an optimal condition that does not need to consider the change in the surrounding environment, but is gradually expanding to the applicable field in the natural environment. Moving object detection in a continuous image obtained from a moving camera in a natural environment is a field that is being studied in computer vision and various applications as the initialization for moving object tracking. In particular, detecting moving objects in a moving camera is one of the most challenging tasks in the field of computer vision because of the background motion. Moving object detection is applied to various applications such as object recognition, robots, surveillance systems and unmanned vehicle systems.

The detection of a moving object in a continuous image obtained from a mobile camera in which background motion occurs is made possible by analyzing the motion in the image. Models for analyzing motion in an image can be classified into parametric models and non-parametric models. The parametric model is used to represent coordinates in three-dimensional space as coordinates on a two-dimensional image plane by orthographic or perspective projection. Although this model provides analytic information of the three-dimensional system, it has a disadvantage that it can be applied only to the analysis of three-dimensional rigid motion. The method using the parameter model is able to interpret relatively accurate movements in the individual image units, but when it is applied to the image sequence in which the image acquisition environment changes rapidly, it requires a lot of calculation to identify the relationship between the images and also the non-rigid movements. There is also a limit to the interpretation of non-rigid motion. On the other hand, the non-parametric model enables intuitive analysis of two-dimensional images and uses the model as an optical-flow based estimation method and a feature-based method, which is represented by a deterministic motion based method. feature based), and block based estimation.

Conventional methods for detecting moving objects as the initialization of object tracking are mainly based on optical flow and feature-based methods. The optical flow method recognizes and tracks the movement of an object by using the space-time vector when the object obtained in the previous frame is located at the next frame through time and space changes. The method proposed by Horn et al. Obtains accurate light flow in the case of moderate brightness in continuous images, but it is difficult to obtain accurate solutions in the case of rapidly changing images. And, iterative calculation is required for each pixel, which requires a lot of computational burden. The method using the optical flow has a disadvantage of slow processing because it is difficult to apply in the outside and the amount of calculation is large because the brightness value of the image is changed by various factors. The method using the feature is a method of separating the object from the background by obtaining different motion information of the moving object and the background from the image by using the feature of the image in the continuous image obtained from the moving camera. The features of the image are the boundaries between the faces, the intersections of lines, the edges of objects, and the maximum / minimum points of curves. In the human visual system, the existence of objects in the background is determined by first distinguishing the objects in the background by the characteristics of the image, and the shape of the separated objects is connected to the feature points in the outer boundary of the object. Is determined by. Determination and tracking of object movement are made possible by identifying the movement relationship of these feature points. The moving object can be detected by obtaining a motion path representing the movement of the feature point over time, and performing clustering on the feature space using the position and the angle and size of the feature vector constituting the feature-based motion path as parameters. . The edge point used as the feature path-based information includes methods such as Smallest Univalue Segment Assimilationg Nucleus (SUSAN), Kanade-Lucas-Tomasi (KLT), Harris (Harris), and the like. This feature-based method is difficult to apply to an image in which feature points do not appear well, and it is difficult to obtain uniform background information in an image structure for detecting a moving object having a constant background movement.

In the hardware system construction of moving object tracking, the detection of moving objects requires a system with a simple algorithm at low cost. In addition, in order to overcome the problems of the existing methods, this paper uses a method that integrates a cumulative double difference and motion estimation technique. We propose an algorithm for automatically detecting omnidirectional moving objects in an image. The cumulative difference image is an image that detects the movement of an intermediate frame using three consecutive frames. The cumulative difference image is used to find a candidate region where a moving object exists by removing a background having a uniform texture from the continuous image. The motion estimation technique is used to remove the background by using the characteristics of the fast change of the background in the continuous image acquired by the mobile camera. By integrating the cumulative difference image and the motion estimation technique, the candidate region found is estimated in block units from consecutive images, and the moving object is detected while eliminating the portions of the estimated motion that are severely changed.

1 is a flowchart illustrating a conventional moving object detection method. The conventional moving object detection method includes: manually selecting an object to be tracked in a previous frame and leaving it as a template (S101); Dividing an object from a background by using a template extracted from a previous frame in a next frame (S102); And updating the object having the highest similarity with the object previously tracked through the matching process among the divided objects (S103).

FIG. 2 is a block diagram illustrating a conventional moving object detection system. In the conventional moving object detection system, each histogram receives an input image I _k and a previous tracking model M _{k-1 of a k-} th frame. A histogram back projection unit (HBP) 210 for acquiring a symbol, recognizing a pattern characterized by a brightness value by performing density filtering, and outputting a result value Bc _k ; An area RM that is likely to have a tracking object obtained in the kth frame by receiving the output signal Bc _k of the histogram back projection unit HBP and the input image I _{k of the} kth frame, and performing a histogram intersection process. _k ) histogram intersection section 220 for obtaining; XY-projection that receives the output signal RM _k of the histogram intersection unit 220 and the input image I _k of the k-th frame, thereby generating a tracking model M _k that detects the outermost part of the object. Part 230; A delay unit 240 for delaying the output signal of the XY-projection unit 230; And an adder 250 that adds an output signal of the delay unit 240 and a model M ₀ of the initially detected moving object, and then outputs the histogram back projection unit HBP 210.

Referring to the operation of the conventional moving object detection system described above is as follows. The moving object to be tracked is given by the system user. A moving object of a given tracking object is used as a template to track an object in a continuous input image. Histogram back projection is used for moving object segmentation in the input image, histogram intersection as a model for matching, and XY-projection is used for accurate object identification and extraction. First, histogram back projection is a useful method for segmenting objects in an image and finding the location of the object to be tracked. The histogram of the input image I is defined as H _I (j), and the histogram of the model object is defined as H _M (j). H _R (j) is the histogram divided by H _M (j) by H _I (j). The image I _B back-projected by H _R (j) can be expressed by the following equation.

The object is likely to be near the highest value of the back projected image I _B. Density filtering was used to get more accurate results. D ^r is defined as a circular mask having a radius r and is represented by the following equation.

Here, r is defined as half the diagonal length of the object image.

In addition, histogram intersection is a kind of classical pattern recognition as a method for histogram matching. This method uses each brightness value in the histogram as a feature point. The histogram intersection is used to determine whether the object is the same as the object to be recognized. The larger the result, the more similar to the object to be recognized. The histogram intersection may be defined as in the following equation.

이고, 모델 영상의 히스토그램은 H_M이다. 정합후보 영상과 모델 영상은 동일한 n개의 밝기값을 갖고 있다. 모델 영상의 히스토그램과 정합후보 영상의 히스토그램 인터섹션의 결과는 정합흐보 영상에서 동일한 밝기 값을 갖는 화소에 대응되는 모델 영상의 화소의 갯수이다.Here, a matching candidate image having the same size as a model image having one point i of the input image I as a center point is made. The histogram of the matched candidate image is

The histogram of the model image is H _M. The matched candidate image and the model image have the same n brightness values. The result of the histogram intersection of the model image and the matched candidate image is the number of pixels of the model image corresponding to the pixels having the same brightness value in the matched image.

Normalize histogram intersection values to have matching values between 0 and 1. A normalized value may be expressed as follows by applying the number of pixels corresponding to the histogram of the model image.

The result of the histogram intersection does not decrease even when the division with the background is not easy. It is not easy to divide the object perfectly in the background, but if the pixels inside the model have the same brightness, the match will be large. If the number of pixels having a specific brightness value in the object to be matched is smaller than the number of pixels having the same value in the model image, the matching value is increased.

On the other hand, the histogram back projection is used to find a position where an object is more likely to be present. Applying the histogram intersection with the found position as the matching candidate point, we obtain the area containing the object to be matched. Performing an XY-projection with this information can find the outermost part of the object. Histogram intersection is performed to perform XY-projection, and vertical and horizontal edges are obtained for the image in the obtained region. The vertical edge images V (x, y, k) and the horizontal edge images H (x, y, k) are obtained using brightness gradients. The horizontal and vertical edges are used to project the x and y axes, and the result obtained by projecting the vertical edges to the x axis is left as u. The result of the projection of the horizontal edge on the y-axis is set to v. The following equation is used.

In order to remove noise and extract objects, a threshold value was set and set, and the threshold value was determined to be 1 / 2.5 times the largest value projected in each direction. The threshold value is expressed by the following equation.

Project the vertical edges in the x-axis direction and examine the values from the left and right ends toward the center. If it is larger than the threshold value during the investigation, it is determined that the left and right outside of the object is found and stops. Similarly, the horizontal edge is processed to extract the object.

In addition, the moving object tracking system described above efficiently tracks objects in a continuous image in which the background is changed by using histogram projection. On the other hand, a method of automatically detecting a tracking target moving object corresponding to initialization for object tracking has not been solved. This is because it is difficult to automatically separate the moving object to be tracked from the background in the continuous image obtained from the moving camera which causes the background motion. In the existing research, tracking initialization consists of manually finding the tracking object in the input image and saving it as a template.

That is, the tracking system for real-time processing should be fast and the moving object can be easily tracked even in the changing background. In this situation, the detection of the moving object on the moving camera reduces and moves the detection area excluding the background. Although the candidate area needs to be set in consideration of the feature of the object, the optical flow-based estimation method and the feature-based estimation method do not accommodate the fast processing speed required by such a moving object detection system, and do not efficiently remove the changing background. There is this.

In order to solve the above problems, the present invention integrates a cumulative difference image and a motion estimation technique to automatically detect a moving object in a continuous image obtained by a moving camera, thereby solving the problem of initializing a conventional moving object tracking system at low cost. It is an object of the present invention to provide a moving object detection system and method for solving the automation of the tracking system.

In order to achieve the above object, the moving object detection system of the present invention receives a sequence image frame sequentially, acquires a cumulative difference image between frames, and generates motion image data by performing a morphological filtering process with the cumulative difference image. A motion information acquisition unit; And a final candidate block detector which receives the sequence image frame and the motion image data, estimates motion through a block matching process, and detects a final candidate block area according to the estimated result.

In addition, the moving object detection method of the present invention to achieve the above object, the step of receiving a continuous frame; Comparing the successive frames and detecting a moving region using a cumulative difference image; Reducing a candidate area in which a moving object is present by removing a background having a uniform texture in the moving area; Receiving a series of frames, performing block matching with reference to a candidate region in which the moving object will exist, and as a result, obtaining a block position where the matching is best and a displacement of a motion vector according to the matching; And comparing the displacement of the motion vector with an allowable boundary value to gradually remove the background by removing blocks having a large displacement to obtain a final candidate region in which a moving object exists.

Hereinafter, the most preferred embodiments of the present invention will be described with reference to the accompanying drawings so that those skilled in the art can easily implement the technical idea of the present invention. .

First, FIG. 3 is a block diagram illustrating a moving object detection system according to an exemplary embodiment of the present invention. The moving object detection system of the present invention includes a motion information acquisition unit 310 and a final candidate block detection unit 330. do.

The motion information acquisition unit 310 sequentially receives the sequence image frame I _k , obtains a cumulative difference image between frames, and performs a morphological filtering process with the cumulative difference image to perform motion image data I _DD . It plays a role in generating. Here, the motion information acquisition unit 310 will be described in detail as follows.

The first delay unit 311 mounted in the motion information acquisition unit 310 plays a role of outputting the delayed frame I _k-1 after receiving and delaying the sequence image frame I _k .

In addition, the second delay unit 312 mounted in the motion information acquisition unit 310 receives a frame I _k-1 from the first delay unit 311 and delays the delayed frame I _{k. -2} ) outputs

Meanwhile, the first difference image generator 313 mounted in the motion information acquirer 310 may output the frame I _k-1 output from the sequence image frame I _k and the first delay unit 311. ) To generate a difference image.

In addition, the second difference image generator 314 mounted in the motion information acquisition unit 310 may include the frame I _k-1 output from the first delay unit 311 and the second delay unit 312. ) Generates a difference image by receiving the output frame I _k-2 .

Meanwhile, the first binarization unit 315 mounted in the motion information acquisition unit 310 receives a difference image output from the first difference image generation unit 313 and converts the difference image into binary data.

In addition, the second binarization unit 316 mounted in the motion information acquisition unit 310 receives a difference image output from the second difference image generation unit 314 and converts the difference image into binary data.

The AND operator 317 mounted in the motion information acquirer 310 receives an AND operation by receiving data output from the first binarizer 315 and data output from the second binarizer 316. And then output the data as a cumulative difference image.

In addition, the adder 318 mounted in the motion information acquisition unit 310 receives and adds the cumulative difference image and the delayed motion image output from the AND operator 317 and outputs the received difference.

Meanwhile, the third delay unit 319 mounted in the motion information acquisition unit 310 generates the delayed motion image by delaying the data output from the adder 318 and generates the delayed motion image. The output unit 318 serves to output.

In addition, the morphological analysis unit 321 mounted in the motion information acquisition unit 310 receives the data output from the adder 318 to reflect expansion and erosion by a morphological filter and then the motion image data. Outputs as (I _DD ).

Meanwhile, the final candidate block detector 330 receives the sequence image frame I _k and the motion image data I _DD output from the motion information acquirer 310, and estimates motion through a block matching process. And detects the final candidate block area BR _k according to the estimated result. Here, the final candidate block detector 330 will be described in detail as follows.

The fourth delay unit 331 mounted in the final candidate block detector 330 receives and delays the sequence image frame I _k and outputs the delayed image.

In addition, the block initialization unit 332 mounted in the final candidate block detector 330 may perform the motion image data I _DD having cumulative motion information from which a background part having a uniform texture is removed using a cumulative difference image technique. ), The motion image data I _DD is divided into block units having a predetermined size to remove blocks that do not have motion information in the block, and to output the remaining blocks.

Meanwhile, the block matching unit 333 mounted in the final candidate block detector 330 inputs a series of frames including the sequence image frame I _k and the frames output from the fourth delay unit 331. In this case, the preceding frame finds the best matching position on the subsequent frame, and outputs the block position and the displacement of the motion vector according to the matching.

Also, the block region updater 334 mounted in the final candidate block detector 330 receives the displacement of the block position and the motion vector from the block matcher 333, and permits the displacement of the motion vector. Backgrounds are gradually eliminated by removing blocks having a large displacement compared to the boundary value, and thereby finding and outputting the final candidate region BR _k in which the moving object exists.

Meanwhile, the fifth delay unit 335 mounted in the final candidate block detector 330 receives and delays the final candidate region BR _k output from the block region updater 334 and outputs the delayed final candidate region BR _k . Do it.

In addition, the switching unit 336 mounted in the final candidate block detector 330 may output data output from the block initialization unit 332 or data output from the fifth delay unit 335. 333 serves as one of the series of frames.

Referring to the operation of the moving object detection system according to an embodiment of the present invention described above are as follows.

First, the motion information acquisition unit 310 detects a region having a motion during a continuous frame by using a cumulative difference image, and removes a background having a uniform texture to reduce a candidate region in which a moving object is to exist. Thereafter, the final candidate block detector 330 estimates motion in blocks by a motion estimation method and estimates a candidate area found by using a cumulative difference image acquired in a continuous frame, and then deviates from a constant size change. Eliminate background completely by sequentially removing. Then, the moving object is detected in the final candidate area using XY-projection. In the detection of omnidirectional moving objects integrating the cumulative difference image and the motion estimation technique, the cumulative difference image efficiently removes the portion having uniform texture, which takes up a large portion of the background of the image, and acquires the motion information of the image. Find a candidate region that exists. The motion estimation technique detects a moving object by removing a background using different motion information of the moving object and the background. The operation of such a system will be described in detail as follows.

First, the images acquired in the natural environment occupy a large part of the background having a uniform texture such as sky and asphalt road. The difference imaging technique removes such components having uniform texture from two consecutive images and makes it possible to detect only moving regions. The cumulative difference image technique efficiently removes a background region having a uniform texture from a continuous image, obtains motion information in the image, and accumulates motion information in the region.

The formula for making the difference image is as follows.

The cumulative difference image is obtained by performing a logical AND with pixels belonging to two consecutive difference images. The formula for making the cumulative difference image is as follows.

Where T represents a threshold. A morphological filter is performed to remove noise on the obtained cumulative difference image. Expansion

는 x만큼 움직인 S(Structuring element) 집합이다.Expressed as here,

Is a set of S (Structuring elements) moved by x.

Expressed as The results of applying the applied morphological filter are as follows.

4A to 4C illustrate a morphological filter after removing a background having a uniform texture using a cumulative difference image in a continuous frame using a motion information acquisition unit 310, detecting and accumulating a moving area. 4A is an exemplary view showing a cumulative difference image using three initial frames, FIG. 4B is an exemplary view showing an image accumulated with motion information for nine consecutive frames, and FIG. 4C is a shape. This is an exemplary view showing an image from which noise is removed using a scientific filter.

In addition, the block-based motion estimation technique applied to the final candidate block detector 330 is a method widely used for video compression for estimating the motion between successive frames I _k-1 and I _k . In general, since a video has a lot of overlapping components in space and time, a high compression ratio cannot be obtained by spatial processing alone. Motion compensation coding is performed to remove temporal redundancy to increase compression efficiency. In the data compression method using the same, motion estimation and compensation are performed using a previous frame, and then the difference image between the original image and the image compensated by the estimated motion vector is encoded. Block-based algorithms, such as block-based motion estimation techniques, require less time for motion estimation and easier hardware implementation than pel recursive algorithms (PRA), such as optical flow-based or feature-based estimation methods. Widely used for motion compensation coding. This method has been applied to video coding standards such as H.261, H.263, MPEG-1, MPEG-2, etc.

Estimation of the individual motion vectors with the motion estimation on a block-by-block basis is usually carried out in the process of matching the block by comparing the entire region in advance inside the defined search window of the frame each block and the frame (I _k) of (I _k-1) do. This has the disadvantage that it takes a long time to execute. This reduces the execution time by using a normalized cross-correlation coefficient having a hierarchical structure.

First, the final candidate block detector 330 can be largely divided into the details of the block initialization unit 332, the block matching unit 333, and the block region updater 334. All processes are performed on the reduced image of the upper layer by the pyramid structure, and the result data is projected to the lower layer to have the result value for the original image. The block initialization unit 332 divides the image I _DD having accumulated motion information from which a background portion having a uniform texture is removed by using a cumulative difference image technique into blocks of a predetermined size and does not have motion information in a block. Remove the blocks. The remaining blocks are then subjected to block matching in certain continuous frames. In addition, the block matching unit 333 performs a process of finding a position where the blocks on the previous frame are matched best on the current frame. Then, after block matching, the block area updater 334 compares the displacement of the motion vectors of the obtained blocks with the tongue boundary value, and gradually removes the background by removing the blocks having large displacements, so that the final object having the moving object exists. The process of finding candidate regions is performed. Referring to FIG. 5, the input image I _{k of} the k th frame and the positions BP _k of the blocks in the k th frame after block matching are shown. In addition, the displacement d of the block motion vector and the candidate region BR _{k in} which the moving object detected by the block will be shown are also shown.

Estimation of motion by block matching divides past frame among two frames non-overlapping into N × N blocks, and divides each N × N block into current frame based on the same point as this block. It is to find the point where N × N block matches best. At this time, it is assumed that all points in the block have the same movement. As shown in FIG. 5, the maximum displacement that can be estimated depends on the size of the search window denoted by w. The block matching algorithm that finds a part having a pure brightness value between pixels does not use an image feature, and thus has an advantage that it can be applied to all kinds of images regardless of whether or not the image is characteristic.

A measure of the degree of match between two blocks that can be matched to each other is the mean absolute difference (MAD), mean squared difference (MSD), and normalized cross-correlation coefficient. ; NCC) method is widely used. MAD, MSD and NCC are each represented by the following formula.

는 현재 프레임의 블록 내 화소들의 밝기 평균값을,

는 이전 프레임의 블록 내 화소들의 밝기 평균값을 나타낸다. 각 탐색점에 대해 MAD와 MSD는 그 값이 최소가 되는 (u,v)를 모션 벡터로 정의하고, NCC는 그 값이 최대하 되는 (u,v)를 모션 벡터로 정의한다. 세 가지 척도는 정확성과 계산량의 장단점이 서로 맞물려 있다. NCC는 세가지 척도 중에서 가장 정확한 모션 벡터를 탐색하지만 계산량이 가장 많다. 그리고, MAD는 계산량이 적고 하드웨어 구현이 용이하여 널리 이용되고 있지만 세가지 중에서 정확성이 가장 떨어진다.Where I _k (i, j) is the current frame, I _k-1 (i, j) is the brightness value at the (i, j) position of the previous frame and (u, v) is the search point in the search region. to be.

Is the average brightness value of the pixels in the block of the current frame,

Denotes a brightness average value of the pixels in the block of the previous frame. For each search point, MAD and MSD define (u, v) as the motion vector with the minimum value, and NCC defines (u, v) as the motion vector with the maximum value. The three measures combine the advantages and disadvantages of accuracy and computation. The NCC searches for the most accurate motion vector of the three measures, but has the highest computational complexity. In addition, MAD is widely used because of its small amount of calculation and easy hardware implementation, but has the lowest accuracy among the three.

According to the present invention, since the matching process is performed on a block-by-block basis only after removing a background having a uniform texture, which occupies a large portion of an image, the number of target blocks is small. In addition, since block matching is performed on a reduced image using a hierarchical structure, a highly accurate normalized cross-correlation coefficient is used as a matching scale.

The motion estimation technique using the hierarchical normalized cross-correlation coefficient removes the block region that escapes the constant size change from the estimated motion vector information for a certain frame, thereby eliminating the background and reducing the candidate region in which the moving object will exist. The displacement d of the block motion vector is as follows.

Here, ρ {} denotes a matching scale function, d _x and d _y denote displacements, W denotes a search region, and x and y denote coordinates of an image. Block region extraction is as follows.

Here, T _d represents a critical range.

6A to 6D divide the motion information image detected using a cumulative difference image method into blocks, find only blocks having motion information among the blocks, and perform motion estimation for 9 consecutive frames to remove the background. It shows the exit process. 6A is an exemplary diagram illustrating a motion information image and block division, FIG. 6B is an exemplary diagram illustrating a block initialized image, FIG. 6C is an exemplary diagram illustrating candidate blocks after 3 frames, and FIG. 6D is 9 frames. It is an exemplary view showing the final candidate blocks after.

If there are two or more moving objects in the image received as input, the final candidate blocks will form corresponding clusters. Detection of multiple moving objects is made possible by performing a labeling process to divide these block clusters.

Each block left by block-based motion estimation may not include the moving object and all parts. In order to accurately detect the moving object, the search window for detecting the moving object is obtained by applying an offset to the rectangular region surrounding the outermost part of the final candidate block cluster. The moving object is detected by performing XY-projection using the vertical and horizontal edges of the moving object within the search and receive.

7A and 7B show the detection of moving objects using XY-projection in the search window area. 7A is an exemplary diagram showing a block cluster indicating a final candidate region, and FIG. 7B is an example showing a search window offset again in an area surrounding the outermost regions of the block clusters and a moving object detected after performing XY-projection. It is also.

8 is a flowchart illustrating a moving object detecting method according to an exemplary embodiment of the present invention.

First, the motion information acquisition unit 310 receives a continuous frame (S801).

Thereafter, the continuous frames are compared to detect a moving region using a cumulative difference image (S802).

Thereafter, the background having the uniform texture is removed to reduce the candidate area in which the moving object is present (S803).

Thereafter, a series of frames are input, block matching is performed by referring to a candidate region where the moving object is to be located, and as a result, a block position having the best matching and a displacement of a motion vector according to the matching are obtained (S804).

Thereafter, the block position and the displacement of the motion vector are input, and the background of the moving object is finally removed by gradually removing the blocks having a large displacement by comparing the displacement of the motion vector with an allowable boundary value. A candidate region is obtained (S805).

The present invention described above is capable of various substitutions, modifications, and changes without departing from the technical spirit of the present invention for those skilled in the art to which the present invention pertains. It is not limited to the drawings shown.

The present invention integrates a cumulative difference image and a motion estimation technique to automatically detect a moving object in a continuous image obtained by a mobile camera, thereby solving the initialization problem of a conventional moving object tracking system at low cost and completing the automation of the tracking system. There is an advantage.

Claims (4)

A motion information acquisition unit configured to sequentially receive a sequence image frame, obtain a cumulative difference image between frames, and generate motion image data by performing a morphological filtering process with the cumulative difference image; And A final candidate block detector for receiving the sequence image frame and the motion image data, estimating motion through a block matching process, and detecting a final candidate block region according to the estimated result; The motion information acquisition unit, A first delay unit receiving and delaying the sequence image frame; A second delay unit configured to receive a frame from the first delay unit and delay the frame; A first difference image generation unit receiving the sequence image frame and the frame output from the first delay unit to generate a difference image; A second difference image generation unit configured to generate a difference image by receiving a frame output from the first delay unit and a frame output from the second delay unit; A plurality of binarizers for receiving the difference image output from the first difference image generator and the difference image output from the second difference image generator and converting the difference image into binary data; An AND operation unit which receives the data output from the plurality of binarization units, performs an AND operation, and outputs the result data as a cumulative difference image; An adder which receives the cumulative difference image and the delayed motion image and adds the accumulated difference image and the delayed motion image; A third delay unit configured to generate the delayed motion image by receiving and delaying the data output from the adder; And A morphological analysis unit that receives the data output from the adder and reflects expansion and erosion by a morphological filter and outputs the motion image data. Moving object detection system comprising a. delete The method of claim 1, wherein the final candidate block detector, A fourth delay unit receiving and delaying the sequence image frame; A block initialization unit dividing the motion image data into block units having a predetermined size, removing blocks not having motion information in the block, and outputting a remaining block; A series of frames including the sequence image frame and the frame output from the fourth delay unit are input, and a block position and a motion vector according to the matching are best matched on a subsequent frame among the preceding frames. Block matching unit for outputting the displacement; A block region update unit which receives the block position and the displacement of the motion vector, compares the displacement of the motion vector with an allowable boundary value, and removes blocks having a large displacement to gradually remove a background to find and output a final candidate region; A fifth delay unit receiving and delaying the final candidate region; And A switching unit for conducting data output from the block initialization unit or data output from the fifth delay unit as one of the frames of the block matching unit; Moving object detection system comprising a. Receiving a continuous frame; Comparing the successive frames and detecting a moving region using a cumulative difference image; Reducing a candidate area in which a moving object is present by removing a background having a uniform texture in the moving area; Receiving a series of frames, performing block matching with reference to a candidate region in which the moving object will exist, and as a result, obtaining a block position where the matching is best and a displacement of a motion vector according to the matching; And Comparing the displacement of the motion vector with an allowable boundary value to gradually remove a background by removing blocks having a large displacement to obtain a final candidate region in which a moving object exists. Moving object detection method comprising a.

Images