JP2012069068A - Image processing apparatus and image processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image processing apparatus and an image processing method that greatly decrease the amount of image data stored in a memory when detecting a moving body by a background difference method.SOLUTION: The image processing apparatus includes an image data input part 1 which acquires a moving image, a histogram generation part 2 which divides individual image frames of the moving image into a plurality of areas, and classifies each of individual divided small images into one of a plurality of gradations of luminance preset by pixels to generate a histogram of frequency, a convolution processing part 3 which convolutes frequencies of individual gradations of the generated histogram into binary values, statistical data calculation part 4 which calculates statistic data by averaging the individual data convoluted by the convolution processing part 3 for a plurality of images in time series, a memory 7 which stores the calculated statistical data, and a moving body area identification part 5 which compares the stored statistic data with convoluted data of a newly input image frame, and identifies whether there is a moving body in each of the individual small images from a change quantity of the data.

Description

  The present invention relates to an image processing apparatus and an image processing method for detecting a moving object from a moving image and outputting the information.

  The installation of surveillance cameras is increasing for crime prevention and understanding of traffic volume. As a method for recognizing a moving body using such a camera, a background subtraction method in which a difference between successive images is obtained and a moving object is extracted has been implemented. For example, in Patent Document 1, a time-series image is accumulated, statistical processing is performed on the data, a background portion is estimated, and a moving body is extracted from a portion other than the background.

JP 9-81714 A

  However, the image processing method of Patent Document 1 requires image storage means for storing an image memory group that is time-series image data in order to achieve the purpose of extracting a moving object. It was necessary to store the image memory group. In addition, in a system in which storage areas are sufficiently prepared, a moving body can be detected by the image processing method described above. However, when incorporated in a device such as a person recognition device, there may be cases where a sufficient storage area cannot be secured. In this case, it is difficult to detect the moving object by the above method.

  Accordingly, in view of such problems, the present invention has an object to provide an image processing apparatus and an image processing method capable of greatly reducing the amount of image data stored in a memory when a moving object is detected by the background difference method. Yes.

In order to solve the above-mentioned problem, an image processing apparatus according to the invention of claim 1 divides an individual image of an acquired moving image into a plurality of areas by acquiring image data input means for acquiring a moving image and dividing the divided individual images. Histogram generation means for generating a histogram of frequency by classifying a small image into any of a plurality of gradations set in advance for each pixel, and the frequency of each gradation of the generated histogram by a predetermined threshold value K Statistical data for calculating statistical data by averaging convolution arithmetic means for convolution to a value (K is an integer of 2 to 4), and individual data convolved by the convolution arithmetic means for a plurality of time-series images A calculation means, a statistical data storage means for accumulating statistical data,
The stored statistical data up to the (M-1) th image is compared with the data obtained by convolving the histogram data into K values for the new Mth image. Mobile object identifying means for identifying the presence or absence of the mobile object, and identification result output means for outputting the identified result.
According to this configuration, in the process of extracting the features from the input image and calculating the probability distribution, the convolution of the probability distribution data calculated from the input image data is held in the memory, so that a small memory capacity is sufficient. . In addition, since the acquired image is divided into a plurality of pieces for discrimination processing, when the number of divisions is increased, the shape of the moving object can be detected.

The invention of claim 2 is characterized in that, in the configuration of claim 1, K = 2.
According to this configuration, the frequency data is accumulated in one of two values, so that the minimum storage capacity is sufficient. Nevertheless, the moving object can also be identified.

An image processing method according to a third aspect of the present invention includes an image data acquisition step for acquiring a moving image, and dividing each acquired image of the moving image into a plurality of areas, and for each of the divided small images for each pixel. A histogram generation step for generating a frequency histogram by classifying the brightness into one of a plurality of gradations set in advance, and the frequency of each gradation of the generated histogram is set to a K value (K is 2 to 4) by a predetermined threshold. A convolution step for convolution to any integer), a statistical data calculation step for calculating statistical data by averaging the individual data convolved in the convolution step with respect to a plurality of time-series images, and the M−1th image The statistical data up to and the new M-th image are compared with the data obtained by convolving the histogram data into K values. And having a mobile identification step of identifying.
According to this method, in the process of extracting the features from the input image and calculating the probability distribution, a further convolution of the probability distribution data calculated from the input image data is held in the memory, so that a small memory capacity is sufficient. . In addition, since the acquired image is divided into a plurality of pieces for discrimination processing, when the number of divisions is increased, the shape of the moving object can be detected.

  According to the present invention, in the process of extracting the features from the input image and calculating the probability distribution, the convolution of the probability distribution data calculated from the input image data is held in the memory, so that a small memory capacity is sufficient. . In addition, since the acquired image is divided into a plurality of pieces for discrimination processing, when the number of divisions is increased, the shape of the moving object can be detected.

1 is a schematic configuration diagram illustrating an example of an image processing apparatus according to the present invention. It is explanatory drawing of a time-sequential image frame sequence. It is explanatory drawing which divided the image frame into L parts. It is explanatory drawing of the histogram of the produced frequency. It is explanatory drawing which shows the flow of the convolution data creation to the Mth frame, (a) is the convolution data of Mth frame, (b) is the convolution data from the 1st frame memorize | stored in the memory to the M-1st frame. (C) shows the statistical data of convolution data from the first frame to the Mth frame. It is explanatory drawing of the process which detects a mobile body from convolution data, (a) is the convolution data of the Mth frame, (b) is the statistical data of the convolution data from the 1st frame to the M-1st frame, (c). Indicates data of the moving object identification result. It is a flowchart which shows the flow of the image processing for a mobile body detection.

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, embodiments of the invention will be described in detail with reference to the drawings. FIG. 1 is a schematic configuration diagram showing an example of an image processing apparatus according to the present invention. As shown in FIG. 1, the image processing apparatus includes an image data input unit 1, a histogram generation unit 2, a convolution processing unit 3, a statistical data calculation unit 4, a moving body region identification unit 5, a memory data storage unit 6, a memory 7, The memory data reading unit 8 and the identification result output unit 9 are configured. These components are constituted by a programmed microprocessor.

  The image data input unit 1 receives video data such as an interphone camera, a digital camera such as a surveillance camera, or a video tape recorder, and outputs image frames in time series. FIG. 2 shows an image frame output at this time.

The histogram generation unit 2 divides the frame into predetermined L pieces as shown in FIG. 3 with respect to the image frame output from the image data input unit 1. For example, it is divided into 20 equal parts vertically and 30 equal parts horizontally to generate all 600 (L = 600) small images of the same size.
Each small image divided in this manner is classified into any of a plurality of gradations set in advance for each pixel, and a frequency histogram is generated. FIG. 4 shows one of the generated histograms. Here, the luminance is divided into 24 gradations, and the luminance for each pixel is classified into any gradation.

  The convolution processing unit 3 further reduces the amount of information by convolving the frequency for each luminance with a certain threshold with respect to the generated histogram. Here, the frequency is divided by one threshold value and convolved into two values. S in FIG. 4 indicates this threshold value. If the frequency is greater than this threshold S, it is folded into “1”, and if it is smaller, it is folded into “0”. The convolution data generated in this way is shown in FIG.

The statistical data calculation unit 4 performs a logical OR operation between the convolution data of the latest image convolved by the convolution processing unit 3 and statistical data (to be described later) up to the previous image read from the memory 7 by the memory data reading unit 8. Average and calculate new statistical data.
FIG. 5 shows an example of this statistical data generation. In FIG. 5, (a) is the convolution data of the Mth frame, (b) is the statistical data of the convolution data from the first frame to the (M-1) th frame stored in the memory 7, and (c) is from the first frame. Statistical data of convolution data up to the Mth frame is shown. The Mth frame is an image frame that is currently input and processed, and the M−1th frame indicates an image frame that has been input and processed immediately before. The first frame is the first image frame that the image data input unit 1 has started to output.
As shown in FIG. 5, the statistical data calculation unit 4 includes the convolution data of the latest image frame (Mth frame) and the predetermined first image frame to the immediately preceding image frame (M−1th frame). The newest statistical data is generated by obtaining a new logical sum with the statistical data generated by calculating the logical sum of the individual convolution data.

  Note that the method of obtaining statistical data from a plurality of data is not limited to this, and a posteriori probability may be calculated and newly used as statistical information. Further, the first image frame may not be the first input image, and may be reset at an arbitrary timing based on time, for example, as the first image frame.

  The memory data storage unit 6 rewrites the statistical data stored in the memory 7 with the latest data by the statistical data calculated by the statistical data calculation unit 4. On the other hand, the memory data reading unit 8 reads statistical data from the memory 7 for comparison with new convolution data generated by the convolution processing unit 3.

The mobile object region identification unit 5 compares the convolution data of the latest image obtained by the convolution processing unit 3 with the statistical data read by the memory data reading unit 8 to identify the mobile object. FIG. 6 is an explanatory diagram of this mobile object extraction process, where (a) is the convolution data of the Mth frame, (b) is the statistical data of the convolution data from the first frame to the M−1th frame, and (c) is the data. Data of the moving object identification result is shown.
The exclusive OR of the convolution data of the Mth frame and the statistical data from the first frame to the M-1 frame is taken, and the presence or absence of the moving object is determined based on the amount of variation. For example, when the brightness is set to all 24 gradations, if there is a change in the convolution data of 3 gradations, specifically “0” changes to “1”, conversely “1” changes to “0”. If there are more than a predetermined number, it is determined that there is a moving object. This determination is performed on each small image, and the moving body area in the image frame is identified.

  The identification result output unit 9 outputs information on the moving body area detected by the moving body area identification unit 5. The output information is displayed on a display or recorded on a recording means such as a hard disk recorder, and is used for detecting a suspicious person, investigating traffic, and grasping the shape of a moving object.

Next, the operation of the image processing apparatus configured as described above will be described. FIG. 7 is a flowchart showing the flow of mobile object detection, and the description will be made based on this flow.
When the histogram generation unit 2 reads the latest M-th frame of a moving image from the image data input unit 1 (S1), the image frame is first divided into L small images (S2). Then, the luminance of each divided small image is obtained for each pixel, and classified into any of a plurality of gradations prepared in advance, and a frequency histogram is created (S3). FIG. 3 shows this histogram, which is generated for each small image.
Further, convolution is performed on the generated histogram (S4). Due to the convolution, the individual luminances are convoluted into binary values as described above.

  Next, the statistical data of the convolution data of the immediately preceding M−1th image is read from the memory 7 (S5), and the logical sum of this statistical data and the Mth convolution data is taken (S6). The new statistical data thus obtained is stored in the memory 7 (S7), and is used for identification of the moving object in the next M + 1th frame.

  Then, the moving body is identified. The convolution data of the Mth frame and the statistical data up to the (M-1) th statistical data are compared (S9), and a numerical value (threshold value) in which the number of fluctuations (fluctuation amount) z of each gradation convoluted into binary values is preset. If it is above, it will be judged that there exists a moving body (S10), and the small image area will be output as a moving body detection area (S11). If this determination is performed for all L small images, the process is terminated.

  In this way, in the process of extracting the features from the input image and calculating the probability distribution, the memory 7 stores the convolution of the probability distribution data calculated from the input image data, so that a small memory capacity is required. In addition, since the acquired image is divided into a plurality of pieces for discrimination processing, when the number of divisions is increased, the shape of the moving object can be detected.

  In the above embodiment, the histogram is convolved into a binary value using one threshold S, but may be convoluted into a ternary value using two threshold values, or may be convoluted into a quaternary value. . If the data is stored in the memory 7 in the binary, the data stored in the memory 7 can be kept to a minimum amount. However, the identification sensitivity of the mobile body is increased by the convolution in the ternary or quaternary.

  1 .. Image data input unit (image data input unit) 2.. Histogram generation unit (histogram generation unit) 3.. Convolution processing unit (convolution operation unit) 4 .. Statistical data calculation unit (statistical data calculation unit) ), 5... Mobile region identification unit (moving body identification unit), 7... Memory (statistical data storage unit), 9... Identification result output unit (identification result output unit)

Claims (3)

Image data input means for acquiring moving images;
Histogram generation that divides individual images of acquired moving images into multiple areas and classifies the divided small images into one of multiple gradations preset for each pixel to generate a frequency histogram Means,
Convolution operation means for convolving the frequency of each gradation of the generated histogram into a K value (K is an integer of 2 to 4) with a predetermined threshold;
Statistical data calculation means for calculating statistical data by averaging the individual data convolved by the convolution operation means for a plurality of time-series images;
Statistical data storage means for accumulating the statistical data;
The stored statistical data up to the (M−1) -th image is compared with the data obtained by convolving the histogram data into K values for the new M-th image. Mobile object identification means for identifying the presence or absence of a mobile object in the image;
An image processing apparatus comprising: an identification result output unit that outputs an identified result. The image processing apparatus according to claim 1, wherein K = 2. An image data acquisition step for acquiring a moving image;
Histogram generation that divides each acquired moving image into multiple areas and classifies each divided small image as one of multiple gray levels preset for each pixel to generate a frequency histogram Steps,
A convolution step of convolving the frequency of each gradation of the generated histogram into a K value (K is an integer of 2 to 4) with a predetermined threshold;
A statistical data calculation step of calculating statistical data by averaging the individual data convolved in the convolution step with respect to a plurality of time-series images;
The statistical data up to the (M−1) -th image is compared with the data obtained by convolving the histogram data into K values for the new M-th image, and the movement within each small image is determined from the amount of change in the data. A moving body identifying step for identifying the presence or absence of a body.

Images