KR100613732B1 - Image Compression System for Digital Video Surveillance and Method for controlling the Same - Google Patents

Abstract

The present invention relates to an image compression system applied when storing or playing back surveillance images. The present invention improves the compression size and the image quality by focusing on the characteristics of the motion image of the surveillance environment rather than the background by further improving the conventional image compression / restoration algorithm and JPEG, H.263, and MPEG (CCITT). .

Surveillance System, Image Compression / Restoration, Discrete Cosine Transform, Entropy Code

Description

Image Compression System for Digital Video Surveillance and Method for controlling the Same}

1 is a schematic configuration diagram of a conventional JPEG image compression method

2 is a schematic configuration diagram of a conventional H.263 / MPEG image compression method

3 is a schematic configuration diagram of a digital video compression system for a video surveillance application according to the present invention;

<Explanation of symbols for main parts of drawing>

19 pixel frame 20 discrete cosine converter

21: Quantizer 22: Frame Buffer

23: write control 24: difference

25: Similarity discriminator 26: Channel controller

27: inter / intra mode determiner 28, 30: DCT coefficient delay buffer

29: entropy encoder

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a video surveillance compression system, and more particularly, to a digital video compression system and a method applied to storing or playing back surveillance video.

Compression methods used in video surveillance system are generally JPEG, MPEG, H.263, which is based on Discrete Cosine Transform (DCT), and these compression methods are mainly used for ideal cosine transform, quantization, Entropy coding is divided into three parts.

The discrete cosine transform method converts an 8x8 pixel unit area into a frequency domain (DCT coefficient). The discrete cosine transform-based image compression method removes the high frequency term and compresses the image. Do it. Quantization is used to remove high frequency terms. Quantization divides a frequency component (DCT coefficient) value into predetermined quantization values according to the component.

Discrete cosine transform method Compression increases the compression rate by removing the high frequency term, and the high frequency term cannot be completely restored to its original state even though it is dequantized during regeneration. However, high-frequency terms are insensitive to the human eye, and sensitive portions (low-frequency terms) can be restored to their original state.

The discrete cosine transform and quantization processes do not actually compress, but transform the data to increase the compression rate, and the process of compressing entropy is the process of entropy encoding. This process assigns codes to the probability of occurrence of coefficients in the data of the discrete cosine transformation and quantization transformation, assigns short codes to coefficients of high probability of occurrence, and assigns long codes to coefficients of low probability of occurrence. You can reduce the average length by placing a new coding scheme in between. This entropy data becomes final compressed data.

FIG. 1 is a general JPEG compression method and data conversion process, which is composed of a pixel frame 1, a discrete cosine converter 2, a quantizer 3, and an entropy encoder 4. As shown in FIG.

The pixel frame data 5 is output as unit pixel block (8x8 pixel) data 6 for every input pixel frame.

The pixel data is input to the discrete cosine converter 2 as 8x8 pixel block data, and the DCT coefficient data 7 is output after the conversion process. This data is input to the quantizer 3, converted into quantized data 8, and then input to the entropy encoder 4. The input data is output as the final compressed data 9.

JPEG can be applied to intra-frame image coding and is used in applications that compress still images.

2 illustrates a H.263 / MPEG compression method, in which a pixel frame 10, a discrete cosine transformer 12, a quantizer 13, an entropy encoder 14, an inverse quantization 16, and an inverse discrete cosine converter ( 18), the reference video frame 17, the motion predictor 15, and the divider 11.

H.263 / MPEG is a method of compressing by using pre-image information on the assumption that neighboring images are very similar.

Compression in this standard way occurs in two major ways: intra mode and inter mode. Intra mode is an image that is completely different from the previous picture, and is compressed in the same way as JPEG and stores a reference image through a recursive loop.

In addition, the inter mode is a case where any part of the current screen is the same as or moves in any part of the previous screen. In this case, the same figure in the previous screen is searched to obtain the movement amount (motion vector) information, and only the information is compressed.

The two modes reproduce the reference image through a recursive looping process of discrete cosine transform, quantization, and entropy encoding, and then output the final compressed data (intra coding), inverse quantization, and inverse discrete cosine transform. Mode) to estimate the motion by comparing the stored reference image with the current image (15), perform discrete cosine transform, quantization, and entropy encoding on the difference image between the reference image and the current image, and output the final compressed data ( Intercoding) and performs compression while selecting and repeating two modes.

The pixel block generation converts the pixel frame 10 data into 8x8 pixel block data, and passes through this process, followed by a divider 11 or a discrete cosine converter 12.

In the case of intra coding, a discrete cosine transform is performed. In the case of inter coding, a cosine or discrete cosine transform is performed.

Inter coding is also coded by dividing into an intra macro block and an inter macro block in macroblock units, and when coding into an intra macro block, a motion vector cannot be obtained. The coefficients subjected to the discrete cosine transform process become final compressed data after the entropy encoding process after the quantization process.

H.263 / MPEG requires a reference image for the entire screen. After performing inverse quantization and inverse discrete cosine transform (IDCT), the reference image is stored in the frame buffer (intra-coding), and the stored reference image is interpolated. In coding, it is compared with the current image and used to obtain a motion vector.

The obtained similar image and the current image are obtained by compressing a difference image at the pixel level. H.263 / MPEG is compression using image similarity and is used in video compression applications.

Conventional standard video compression methods are applicable to surveillance video systems but do not take into account the general features of surveillance video. Unlike general video, the general feature of surveillance video is camera movement, that is, the use of PTZ (pen, tilt, zoom) is very low or fixed. , Objects such as vehicles appear, and in other areas there is little change over time. In connection with the compression function, the similarity between adjacent images is much higher than that of a normal image, and even if a moving object exists, the occurrence frequency is relatively low in time and space.

The conventional video compression system does not consider the prerequisites of such a video surveillance system, and various problems arise when applied to a video surveillance compression system.

An object of the present invention devised to solve the above problems is to implement a discrete cosine transform-based compression scheme, which does not include a motion prediction function in the pixel region, so that the implementation is simple, and the discrete cosine transform of the reference image and the image to be compressed By using the coefficients, the motion detection function and differential image compression are minimized to minimize the length of entropy code to ensure video compression efficiency, and it is possible to convert to standard compressed data such as JPEG, H.263, MPEG by changing the entropy encoding process. The present invention provides a compression system for video surveillance.

The image compression system of the present invention for achieving the above object is a pixel frame for outputting the image of the frame unit as 8x8 pixel block data, a DCT changer for converting the data of the pixel block unit region to the data of the DCT coefficients, DCT A quantizer for dividing coefficient values into predetermined quantization values, a frame buffer for storing DCT coefficients of a reference video, DCT coefficient write control for reading and storing DCT coefficients from the frame buffer according to inter or intra mode, and for inter mode. A similarity discriminator for calculating a difference between the DCT coefficients, a similarity between the reference image and the current image, and calculating auxiliary data for the channel controller; and generating and detecting motion detection bitmap data using the calculated auxiliary data. The channel controller for intra frame control and the code length are added in units of macro blocks. A macro block unit inter / intra mode determiner for selecting a mode having a short sign among inter and intra modes, a DCT coefficient delay buffer for storing DCT coefficients during the mode selection, and a DCT input from the DCT coefficient delay buffer. And a DCT coefficient selector for selecting DCT coefficients under control of the coefficient and the inter / intra mode determiner, and an entropy encoder for entropy encoding.

In addition, the image compression method of the present invention for achieving the above object comprises the steps of reconstructing and outputting the image of the frame unit in the pixel block unit (8x8), and the data of the pixel block unit region to the data of the DCT coefficients Converting, dividing the DCT coefficient values into predetermined quantization values according to the components thereof, storing the DCT coefficients of the quantized reference image, reading the DCT coefficients from the frame buffer according to the inter or intra mode, Storing the data; obtaining a difference value between DCT coefficients of the reference image and the current image; determining similarity between the reference image and the current image using DCT difference coefficients; and using the auxiliary data output by determining the similarity. Predicting the degree of change with the reference image in frame units and determining the size and contour of the moving object on the screen; In the inter frame, the inter code length and the intra code length are obtained in macroblock units, and the coding length is determined by comparing the code lengths (inter / intra) with a smaller code length, and the coding mode is determined in macroblock units. During the inter- and intra mode, the DCT coefficients are stored and the coefficients stored as the entropy input are selected and entropy coded after the inter / intra mode is determined in units of macro blocks.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Figure 3 is a schematic block diagram of the internal functions of the video compression monitoring system of the present invention.

As shown in FIG. 3, the video compression monitoring system according to the present invention outputs data of a pixel frame 19 and an 8x8 pixel block unit area to reconstruct and output an image of a frame unit in a pixel block unit (8x8). A discrete cosine transformer 20 for converting the data to the data of a coefficient), a quantizer 21 for dividing a frequency component (DCT coefficient) value into predetermined quantization values according to the component, and storing a DCT coefficient of the quantized reference image. Reference image discrete cosine transform coefficient frame buffer 22, DCT coefficient write control 23 for reading DCT coefficients from frame buffer in inter mode and DCT coefficients in frame buffer in intra mode, DCT coefficient and current of reference image Discrete cosine transform coefficient differencer (24), which calculates the difference of DCT coefficients of the image, uses the DCT difference coefficient to determine similarity between the reference image and the current image. Using the auxiliary data output from the sex discriminator 25 and the block / macroblock similarity discriminator 25, the degree of change with the reference image is predicted on a frame-by-frame basis and the size and outline of a moving object on the screen are determined. The channel controller 26 obtains the inter code length and the intra code length in macroblock units within one inter frame, and compares the code lengths to determine a coding mode in which a small code length is less. The determiner 27 stores the inter / intra mode DCT coefficients while determining the coding mode in units of macroblocks, and uses a coefficient of discrete cosine transform coefficients using the coefficients stored as entropy inputs after the inter / intra mode is determined in units of macroblocks. The DCT coefficients (inter mode, in) are controlled by the DCT coefficients input from the buffers 28 and 30 and the delay buffer and the inter / intra mode determiner 27. Select a la mode) it is composed of the entropy encoder 29 that the entropy encoding.

The schematic operation of the present invention is input to the discrete cosine transformer 20 after the pixel frame data is converted to the 8x8 pixel block data output by the pixel frame 19. The input data undergoes a discrete cosine transform process and then undergoes a quantization process through the quantizer 21. These three processes (pixel frame, discrete cosine transform, and quantization) are performed in common in inter / intra mode, and the processes performed thereafter depend on the inter / intra mode.

The inter mode compresses the difference image between the reference video DCT coefficient and the current video DCT coefficient. In order to perform the compression in the inter mode, the reference video DCT coefficient should be stored in the frame buffer.

The storage process is performed in the intra mode, and the intra mode stores the data passed through the discrete cosine transducer 20 and the quantizer 21 in the reference image DCT coefficient frame buffer 22 by the DCT coefficient write control 23. While storing in the intra DCT delay buffer 30. The DCT coefficient write control 23 is controlled by the channel controller 26. The reference image DCT coefficients stored in the delay buffer perform entropy encoding in an intra mode.

Intra mode is applied when the screen to be compressed is not similar to the reference screen or the contents are changed due to the change of illumination, the PTZ operation of the camera, etc., and this frame is used as a new reference image and the discrete cosine transform coefficient is used as the reference image information. In this case, the compressed data is generated by entropy encoding the coefficient while replacing the previous reference image information.

In order to select the intra mode, the channel controller 26 outputs an intra coding command and the output coding control command is input to the macro block unit inter / intra mode determiner 27, and the DCT coefficient selector outputs the intra DCT coefficient delay buffer. Select the data.

The selected intra DCT coefficient is subjected to entropy encoding based on the code of the intra mode code table. After performing the intra coding, the mode is switched to the inter mode by the channel controller to perform compression.

The inter mode compresses the reference image DCT coefficient frame data stored in the intra mode and the differential image of the current image DCT coefficient data passed through the discrete cosine transformer 20 and the quantizer 21 and determines the inter / intra in macroblock units. If the code length is small, entropy encoding is selected.

The inter mode determines the inter / intra mode in units of macroblocks, and the current image DCT coefficients passed through the quantizer 21 are delayers 24, inter / intra mode determiners in macroblock units 27, and DCT coefficient delays, respectively. It is input to the buffer 30 (macro block intra mode).

The difference unit 24 calculates a difference value between the reference image DCT coefficient and the current image DCT coefficient, and the difference image is a similarity discriminator 25 with a block / macro block unit reference image, and a macro block unit inter / intra mode determiner ( 27) is input to the inter DCT coefficient delay buffer 28 (macro block inter mode).

Determining the inter / intra in macroblock units is an inter / intra mode determiner 27 in macroblock units, and receives a difference coder 24 and DCT coefficients of the current image, sums the respective code lengths, and has a small code length. Select the mode.

A mode having a smaller code length among the DCT coefficients output from the inter DCT coefficient delay buffer 28 and the intra DCT coefficient delay buffer 30 is selected to undergo an entropy encoding process.

The inter / intra mode decision is largely divided into frame units and macro block units. When the mode is determined in units of frames, the similarity discriminator 25 with the reference image is output to the block / macro unit reference image, and the channel controller outputs the similarity assistance data with the reference image. The inter / intra mode of the frame unit is determined by predicting and determining the size and contour of the moving object on the screen.

In one frame, the macro block unit inter / intra mode determiner 27 selects a mode having a small encoding length in units of macro blocks.

When the inter / intra mode determiner 27 determines the inter / intra mode in units of macroblocks, the operation prediction is not performed, and the input pixel itself is intra-coded, or a corresponding pixel of the reference image (that is, a position pixel whose motion vector = 0). It is determined that the compression ratio is high among the inter codings with the difference with.

The channel controller 26 determines the inter / intra in units of frames, and the macro / block inter / intra mode determiner 27 determines the inter / intra modes in units of macroblocks within one frame during inter coding.

The pixel difference calculation for inter coding requires no motion prediction function in the compression system, and since the discrete cosine transform is a linear function, the same result can be obtained even if the difference is directly obtained in the quantized discrete cosine transform coefficient region.

Therefore, the functions related to the recursive loop (inverse quantization, inverse discrete cosine transform) required for generating a reference video in MPEG, H.263, etc. are not necessary and the implementation is simplified.

The entropy code table to be applied to the inter / intra mode exists separately to ensure the optimal compression ratio for that mode, and the metric calculation to determine the inter / intra mode for each macro block is the amount of final output that appears after simultaneous inter and intra mode entropy encoding. It is to calculate the (entropy code bit number) and guarantee the minimum bit coding (maximum compression rate) for each macro block.

The present invention can increase the compression efficiency compared to an inaccurate determination method using a sum of absolute distance (SAD), a dispersion value, and the like at a pixel level performed by the existing video standard.

In the inter / intra mode, the entropy coding input is a difference from the quantized discrete cosine transform coefficient or the reference discrete cosine transform quantized discrete cosine transform coefficient of the current macroblock. Therefore, it is possible to code conversion into JPEG, MPEG, etc. at the entropy code level, so that standard compressed file can be generated without changing the quality.

As described above, the present invention can be easily implemented by eliminating the recursive loop for motion prediction. Since the final output is simultaneously output with the minimum bit coding after entropy coding for the inter / intra mode is performed, the compression efficiency can be improved.                     

In addition, it is possible to generate compressed files of other compression standards without changing image quality at the entropy code level, and there is an advantage of ensuring high quality image quality because there is no image quality degradation caused by conversion.

Claims (14)

In the video compression system for video surveillance, A pixel frame controller for converting / outputting image of frame unit into 8 × 8 pixel block data; A DCT changer for converting the data in the pixel block unit area into data of the DCT coefficients, a quantizer for dividing the DCT coefficient values into predetermined quantization values, A frame buffer for storing DCT coefficients of the reference video; A DCT coefficient read / write controller for controlling reading or writing of DCT coefficients from the frame buffer in accordance with an inter or intra mode; A differentiator for finding the difference of the DCT coefficients in the inter mode; A similarity discriminator for determining auxiliary data for the channel controller by determining similarity between the reference image and the current image; A channel controller for generating motion detection bitmap data and controlling motion detection intra frame using the calculated auxiliary data; A macro block unit inter / intra mode determiner for summing code lengths in units of macro blocks to select a mode having a shorter sign among inter and intra modes; A DCT coefficient delay buffer for storing DCT coefficients during the mode selection; A DCT coefficient selector for selecting a DCT coefficient under control of the DCT coefficient input from the DCT coefficient delay buffer and the inter / intra mode determiner, and An image compression system comprising an entropy encoder for entropy encoding. The method of claim 1, The inter mode compresses the difference image between the reference image DCT coefficient and the current image DCT coefficient, and uses a differential calculation method in a DCT coefficient region in which a recursive loop is not required, rather than a pixel differential calculation method through a recursive loop. Image Compression System. The method of claim 2, And a reference image DCT coefficient is stored in a frame buffer in order to perform compression in the inter mode. The method of claim 1, The intra mode is a video compression system, characterized in that the data passed through the quantizer is stored in the intra DCT delay buffer while the data stored in the reference video DCT coefficient frame buffer by the DCT coefficient write control. The method of claim 1, The DCT coefficient write control is controlled by a channel controller, wherein the channel controller performs control according to a channel by using auxiliary data output from a similarity discriminator. The method of claim 1, In order to select the intra mode, a coding control command output from a channel controller is input to a macro block unit inter / intra mode determiner, and the video compression system selects output data of the intra DCT coefficient delay buffer from the DCT coefficient selector. . The method of claim 1, The inter mode determines the inter / intra mode in units of macroblocks, and the current video DCT coefficients, which have passed through a quantizer, are input to a difference blocker, an inter / intra mode determiner in units of macroblocks, and a DCT coefficient delay buffer. system. The method of claim 1, And the difference unit calculates a difference value between the reference video DCT coefficient and the current video DCT coefficient. The method of claim 1, The macro block unit inter / intra mode determiner receives the DCT coefficients of the difference and the current image, obtains the sum of the code lengths, and selects a mode having a small code length. The method of claim 1, The inter DCT coefficient delay buffer and the intra DCT coefficient delay buffer store DCT coefficients input to the inter / intra mode determiner. Image compression system, characterized in that the input. The method of claim 1, The inter / intra mode determiner determines that the compression ratio is high among the inter-coding of the input pixel itself or the inter-coding of the reference image with the difference value of the corresponding pixel without performing operation prediction. system. The method of claim 1, And the entropy encoder selects a DCT coefficient input from the inter or intra DCT coefficient delay buffer and a DCT coefficient under control of an inter / intra mode determiner. The method of claim 1, And convert the entropy encoding process into standard compressed data of JPEG, H.263, and MPEG. In the video compression method for video surveillance, Reconstructing and outputting an image in a frame unit in pixel block units (8x8); Converting data of the pixel block unit area into data of a DCT coefficient; Dividing the DCT coefficient value by a predetermined quantization value according to the component thereof, Storing the DCT coefficients of the quantized reference image; Reading and storing DCT coefficients from the frame buffer according to the inter or intra mode; Obtaining a difference value between the DCT coefficients of the reference video and the current video; Determining the similarity between the reference video and the current video using the DCT difference coefficient; Determining the size and contour of a moving object on the screen by predicting the degree of change with a reference image on a frame-by-frame basis by using the auxiliary data output by determining the similarity; Determining an inter-code length and an intra-code length in macroblock units within one inter frame, and comparing the code length to determine a mode having a smaller code length as a coding mode; Storing the inter / intra mode DCT coefficients while determining the coding mode in units of macroblocks, and using the coefficients stored as entropy inputs after the inter / intra mode is determined in units of macroblocks; And entropy-signaling by selecting a DCT coefficient.

Images