TWI476703B - Real-time background modeling method - Google Patents

Description

Instant background modeling method

The present invention relates to a background modeling method, and more particularly to an instant background modeling method that can be used to quickly update a reference background model.

At present, how to detect foreground objects in fixed-camera video surveillance movies is a basic step in many computer vision applications. The general practice is to use the background model (Background Modeling) to learn the background and use the learned The background model is used to perform foreground detection on subsequent input images. In previous studies, many methods for establishing background models were proposed, such as using color and direction information, or using pixel and block information to describe features of the background. In describing these features, the most commonly used methods are Gaussian Mixtures and the use of Local Binary Patterns (LPBs) to quickly build models.

Of course, the use of background models for foreground detection must overcome several common problems: changes in brightness (shadow or glare), moving background (water waves or leaves), background changes (such as long-term background stops in a car). The better you want to overcome these problems, the more computing time you will have to pay for immediate processing. Background models are the basis for many applications, such as face recognition, object tracking, and more. If you don’t get the right prospects from a good background model, the latter identification and tracking will more difficult.

Therefore, how to establish a fast, highly adaptable, and stable model in a fixed background of limited changes is a problem to be solved by the present invention.

In view of the above-mentioned problems of the prior art, one of the objects of the present invention is to provide an instant background modeling method to establish a fast, highly adaptable and stable background model.

According to the purpose of the present invention, an instant background modeling method is provided, which is suitable for a surveillance camera device. The surveillance camera device is provided with an image capture module and a processing module, and the method comprises the following steps: capturing by image The module captures a plurality of reference images; the reference modules are respectively divided into a plurality of reference blocks that do not overlap each other through the processing module; and the processing module calculates a unit reference data according to each reference block of each reference image. And a pair of bit reference materials; the image capturing module captures an instant image; and the processing module divides the instant image into a plurality of non-overlapping plurality of instant blocks according to the cutting manner of each reference image; Calculate one unit of real-time data and one pair of real-time data according to each real-time block; use the processing module to compare the unit-yuan real-time data with each unit-element reference data corresponding to the location, or use the dual-bit real-time data Comparing the two-bit reference materials corresponding to the positions to respectively generate a matching result; and according to each matching by the processing module Fruit selectively performing a data update program to update each of the unit cell of each double-byte references or references.

Preferably, the instant background modeling method of the present invention further comprises the following steps: calculating, by the processing module, the number of one-bit transitions according to each unit reference data of each reference image; Comparing with a smoothing threshold respectively, if the number of bit transitions is less than or equal to the smoothing threshold, the processing module selectively compares the real-time data of the unit and the unit reference data corresponding to the position, if The number of bit transitions is greater than the smoothing threshold, and the processing module selectively uses the dual-bit real-time data to compare with the dual-bit reference materials corresponding to the location.

Preferably, the instant background modeling method of the present invention further comprises the following steps: calculating, by the processing module, a unit-yuan distance value according to each unit-element reference data corresponding to the unit-yuan real-time data and the position, or dividing the double-bit element Each double-bit reference data corresponding to the real-time data and the location respectively calculates a double-bit distance value; and the processing module performs a pairwise distance value according to each unit-element distance value or each double-distance distance value, respectively, to Generating a matching result; and if one of the unit distance distance values or one of the two bit distance values is less than the distance threshold in the matching result, the processing module executes a weight update procedure, and each unit element When the distance value or each double bit distance value is greater than the distance threshold, the data update procedure is performed by the processing module.

Preferably, the weight update procedure comprises the steps of: updating one unit weight of each unit reference material by the processing module A double value weight value that is a double value or one of the double bit references.

Preferably, the data update program further comprises the steps of: sorting the unit weight values of each unit reference material by the processing module, or sorting the double bit weight values of each of the double-bit reference materials; Substituting the unit meta-reference data with the smallest value among the unit weight values as the unit-yuan real-time data through the processing module, or replacing the double-bit reference material with the minimum value among the double-bit weight values as the dual-bit instant data Data; and the processing module replaces the unit weight value of the minimum value or the double-weight weight value of the minimum value with an initial weight value.

Preferably, each unit reference material and unit time data meet the following conditions: Wherein, TH _smooth is a preset threshold, x _ij is the pixel value of the position (i, j) of each reference block or instant block, and m is the average value of each reference block or instant block. a _ij is a unit value of the position (i, j) of each unit element reference material or unit time data.

Preferably, each double-bit reference material and the dual-bit real-time data satisfy the following conditions: Wherein, TH _smooth is a preset threshold, x _ij is the pixel value of the position (i, j) of each reference block or instant block, and m is the average value of each reference block or instant block. Hm is the average of the average value calculated when the unit value of the unit reference data of each reference block is 1 or the unit value of the real-time data of the real-time block is 1. Lm is the average of the average value calculated when the unit value of the unit reference data of each reference block is 0 or the unit value of the real-time data of the real-time block is 0. b _ij is a double-bit value of the position (i, j) of each double-bit reference material or double-bit real-time data.

Preferably, the instant background modeling method of the present invention further comprises the following steps: calculating, by the processing module, a three-dimensional reference material according to each unit reference material and each double-bit reference data, and according to the unit The real-time data and the dual-bit real-time data are used to calculate a three-dimensional real-time data; and the processing module is used to compare the three-dimensional real-time data with the three-dimensional reference materials corresponding to the positions to respectively generate matching results, and Each three-bit reference material is selectively updated according to each matching result.

Preferably, each of the three-bit reference material and the three-dimensional instant data satisfy the following conditions: Wherein, TH _smooth is a preset threshold, x _ij is the pixel value of the position (i, j) of each reference block or instant block, and m is the average value of each reference block or instant block. Hm is the average of the average value calculated when the unit value of the unit reference data of each reference block is 1 or the unit value of the real-time data of the real-time block is 1. Lm is the average of the average value calculated when the unit value of the unit reference data of each reference block is 0 or the unit value of the real-time data of the real-time block is 0. Hhm is calculated when the double-bit value of the double-bit reference data of each reference block is 11 or the double-bit value of the real-time data of the real-time block is 11 The average number, hlm is the average of the double-bit real-time data of the average of the double-bit reference data of each reference block or the double-bit real-time data of the real-time block is 10 The average calculated by the time, lhm is the double-bit value of the double-bit reference material of each reference block is 01 The average calculated by the time or the double-bit real-time data of the real-time block is the average calculated by the double-bit value of 01, and the llm is the double position of the double-bit reference data of each reference block. The average value calculated by the ternary value of 00 or the average of the double-bit real-time data of the real-time block is 00, and c _ij is the three-dimensional reference material or three digits. One of the three-digit values of the location (i, j) of the meta-instant data.

Preferably, the instant background modeling method of the present invention further comprises the following steps: when the processing module compares the three-dimensional real-time data with the three-dimensional reference data corresponding to the position, the three-digit calculation is respectively calculated. a distance value, and comparing each three-bit distance value with a distance threshold to generate a matching result; and if the matching result, one of the three-dimensional distance values is less than the distance threshold, then A weight update procedure is executed by the processing module, and when the three-bit distance value is greater than the distance threshold, the data update procedure is performed by the processing module.

As described above, the instant background modeling method of the present invention may have one or more of the following advantages:

(1) The instant background modeling method can replace the calculation of the more complex variation number by calculating the number of bit transitions, so that the method of the present invention can quickly update the background model, thereby reducing the computation time of the method.

(2) The instant background modeling method can make the method have high tolerance for brightness change by executing the data update program and the weight update program.

(3) This instant background modeling method can improve the correctness of the foreground information from the background model by applying the multi-bit mode to each block.

The technical features, contents, and advantages of the present invention, as well as the advantages thereof, can be understood by the present inventors, and the present invention will be described in detail with reference to the accompanying drawings. The subject matter is only for the purpose of illustration and description. It is not intended to be a true proportion and precise configuration after the implementation of the present invention. Therefore, the scope and configuration relationship of the attached drawings should not be interpreted or limited. First described.

Please refer to FIG. 1 , which is a flow chart of the instant background modeling method of the present invention.

In step S11, the plurality of reference images are captured by the image capturing module.

In step S12, each of the reference images is divided into a plurality of reference blocks that do not overlap each other via the processing module.

In step S13, the processing module calculates a unit cell reference data and a double bit reference data according to each of the reference blocks of each of the reference images.

In step S14, an image is captured by the image capturing module.

In step S15, the real-time image is divided into a plurality of non-overlapping instant blocks according to the cutting manner of each of the reference images.

In step S16, the processing module calculates a unitary real-time data and a double-bit real-time data according to each of the real-time blocks.

In step S17, the processing unit is used to compare the real-time data of the unit and the unit reference data corresponding to the position, or the double-bit real-time data and the position corresponding to the double-bit reference material are performed. Compare to produce a matching result.

In step S18, the processing module selectively performs a data update procedure according to each matching result to update each of the unit cell reference materials or each of the dual-bit reference materials.

Please refer to FIG. 2, which is a schematic diagram of a first embodiment of the instant background modeling method of the present invention. As shown in the figure, after the monitoring camera captures a plurality of reference images, the plurality of reference images are first divided into a plurality of reference blocks that do not overlap each other. The reference block (a) in the figure is one of a plurality of reference blocks. Then, the processing unit uses the following equation to calculate the unit reference data (b) according to the reference block (a): Wherein, TH _smooth is a preset threshold, x _ij is the pixel value of the position (i, j) of each reference block or instant block, and m is the average value of each reference block or instant block. a _ij is a unit value of the position (i, j) of each unit element reference material or unit time data.

In more detail, the average value m of the reference block (a) in Fig. 2 is 99.25, and the processing module is respectively the value and the flat value in the reference block (a). The mean m is added to the preset threshold for comparison. If the value in the reference block (a) is less than the average value plus the preset threshold, the value of the unit reference data is 0. If the value in the reference block (a) is greater than or equal to the average plus the preset At the threshold, the value of the unit reference is 1. Thus, when each pixel value in the reference block (a) is calculated by the above equation, the unit cell reference data (b) can be calculated.

After the processing module calculates the unit reference data (b), the processing module calculates the double-bit reference material (b) according to the reference block (a) and the unit reference data (b) using the following equation: Wherein, TH _smooth is a preset threshold, x _ij is the pixel value of the position (i, j) of each reference block or instant block, and m is the average value of each reference block or instant block. Hm is the average of the average value calculated when the unit value of the unit reference data of each reference block is 1 or the unit value of the real-time data of the real-time block is 1. Lm is the average of the average value calculated when the unit value of the unit reference data of each reference block is 0 or the unit value of the real-time data of the real-time block is 0. b _ij is a double-bit value of the position (i, j) of each double-bit reference material or double-bit real-time data.

In more detail, the processing module calculates the average number hm based on the pixel value of the unit cell reference data having a unit value of 1, and its value is 108.86. The processing module then calculates an average number lm based on the pixel value of the unit element reference data with a unit value of 0, and the value is 91.77. After calculating hm and lm, the processing module compares the value in reference block (a) with hm plus TH _smooth , m plus TH _smooth and lm plus TH _smooth . If the value in the reference block (a) is less than lm plus TH _smooth , the value of the double-bit reference material is 00. If the value in reference block (a) is between lm plus TH _smooth and m plus TH _smooth , the value of the double-bit reference is 01. If the value in reference block (a) is between m plus TH _smooth and hm plus TH _smooth , the value of the double-bit reference is 10. If the value in the reference block (a) is greater than or equal to hm plus TH _smooth , the value of the double-bit reference material is 11. Thus, when each pixel value in the reference block (a) is calculated by the above equation, the double-bit reference material (c) can be calculated.

It is worth noting that the calculation of the unit reference data and the dual-bit reference data utilizes the information of the entire block, so that the correct rate of the prospective information can be improved, and the change of the brightness is highly tolerated. In addition, since the present invention requires only a simple calculation, the calculation time of the method is short and can be applied to an instant surveillance camera.

Please refer to FIG. 3, which is a schematic diagram of a second embodiment of the instant background modeling method of the present invention. As shown in the figure, the processing module divides the plurality of reference images into a plurality of reference blocks (a) that do not overlap each other and calculates a unit reference material (b). The processing module then calculates the double-bit reference material (c) according to the reference block (a) and the unit reference material (b). The detailed calculation process of the unit element reference material (b) and the double-bit reference material (c) is the same as that of the first embodiment, and the present invention is not described herein. After the processing module calculates the double-bit reference data (c), the processing module calculates the three-dimensional reference data according to the reference block (a) and the double-bit reference data (c) using the following equation (d). : Wherein, TH _smooth is a preset threshold, x _ij is the pixel value of the position (i, j) of each reference block or instant block, and m is the average value of each reference block or instant block. Hm is the average of the average value calculated when the unit value of the unit reference data of each reference block is 1 or the unit value of the real-time data of the real-time block is 1. Lm is the average of the average value calculated when the unit value of the unit reference data of each reference block is 0 or the unit value of the real-time data of the real-time block is 0. Hhm is calculated when the double-bit value of the double-bit reference data of each reference block is 11 or the double-bit value of the real-time data of the real-time block is 11 The average number, hlm is the average of the double-bit real-time data of the average of the double-bit reference data of each reference block or the double-bit real-time data of the real-time block is 10 The average calculated by the time, lhm is the double-bit value of the double-bit reference material of each reference block is 01 The average calculated by the time or the double-bit real-time data of the real-time block is the average calculated by the double-bit value of 01, and the llm is the double position of the double-bit reference data of each reference block. The average value calculated by the ternary value of 00 or the average of the double-bit real-time data of the real-time block is 00, and c _ij is the three-dimensional reference material or three digits. One of the three-digit values of the location (i, j) of the meta-instant data.

In more detail, the processing module calculates the average number hm based on the pixel value of the unit cell reference data having a unit value of 1, and its value is 108.86. The processing module then calculates an average number lm based on the pixel value of the unit element reference data with a unit value of 0, and the value is 91.77. The processing module then calculates an average number hhm based on the pixel value of the double-bit reference material with a double bit value of 11, which is 115. The processing module then calculates the average number hlm based on the pixel value of the double-bit reference material with a double bit value of 10, and the value is 104.25. The processing module then calculates an average value of lhm based on the pixel value of the double-bit reference material with a double bit value of 01, and the value is 95.4. The processing module then calculates an average of llm based on the pixel value of the double-bit reference data with a double bit value of 00, which is 87.25.

After calculating hm, lm, hhm, hlm, lhm, and llm, the processing module is respectively added to the reference block (a) and hhm plus TH _smooth , hm plus TH _smooth , hlm plus TH _smooth , m plus TH _smooth , lhm plus TH _smooth , lm plus TH _smooth and llm plus TH _smooth for comparison. If the value in reference block (a) is greater than hhm plus TH _smooth , the value of the three-bit reference material is 111. If the value in reference block (a) is between hm plus TH _smooth and hhm plus TH _smooth , the value of the three-bit reference is 110. If the value in reference block (a) is between hlm plus TH _smooth and hm plus TH _smooth , the value of the three-bit reference is 101. If the value in reference block (a) is between m plus TH _smooth and hlm plus TH _smooth , the value of the three-bit reference is 100. If the value in reference block (a) is between lhm plus TH _smooth and m plus TH _smooth , the value of the three-bit reference is 011. If the value in reference block (a) is between lm plus TH _smooth and lhm plus TH _smooth , the value of the three-bit reference is 010. If the value in reference block (a) is between llm plus TH _smooth and lm plus TH _smooth , the value of the three-bit reference is 001. If the value in the reference block (a) is less than or equal to llm plus TH _smooth , the value of the three-bit reference material is 000. Thus, when each pixel value in the reference block (a) is calculated by the above equation, the double-bit reference material (c) can be calculated.

It should be noted that the unit cell, double bit and three bit mode of the present invention can be further extended to four-bit, five-bit, multi-bit mode and the like according to requirements.

Further, the present invention is explained in more detail as follows: After the image capturing device intercepts an image, the frame is first divided into blocks having a size of n × n pixels that do not overlap. In each block, the mean m is calculated and satisfies the following equation: Where x _ij is the pixel value in the (i, j) position of the block.

The output value of each image block is a binary bitmap (Bitmap, or simply BM) having the same size as the block. This binary bit map is generated by the following equation: Where a _ij represents the bit value of the position (i, j) of the binary bit map, the bit value 1 indicates that the pixel value of the block is greater than the average value m, and the bit value 0 indicates the block. The pixel value is less than the average value m. Therefore, the combination of binary bit maps is the texture descriptor of the input frame.

Initially, each input frame is partitioned into non-overlapping blocks, and each block is converted to a binary bit map according to the texture descriptors presented above. It should be understood that if the pixels belong to a smooth block, the corresponding texture descriptions may not be robust enough because the pixel values may be very close to the average. Therefore, the binary bit map of the smooth block may have an unstable 0 or 1 bit value. In order to solve this problem, the equation for generating a slightly improved binary bit map of the present invention is the following equation: According to the experimental observation of the present invention, the value of TH _smooth can be 8.

In addition, the image captured in the real-time surveillance camera is a color image, so each block should have a binary bit map of red color frequency, green color frequency and blue color frequency. In the present invention, for convenience of description, each block is represented only as one binary bit map. The method of the present invention can be intuitively extended to three binary bit maps in each block. Figure 4 shows the image produced by the texture descriptor. The texture of this image is clearly Presented to prove the validity of this texture descriptor.

We now consider how to use feature vectors to construct a background model. The background model for each block contains K weighted binary bit maps (BM ₁ , BM ₂ , ..., BM _k ), where each weight is between 0 and 1, and K The sum of the weight values is 1. The weight of the Kth weighted binary bit map is expressed as w _k . When a new block BM _{new is} input, BM _new is compared with K binary bit maps using the Hamming distance equation as described below: Where m ranges between 1 and K.

In the background model, if ( BM _new , BM _m ) is less than a predetermined threshold, then the block BM _new is matched to BM _m and triggers execution of a weight update procedure; otherwise, BM _new is a foreground block and triggers execution of a data update program. Since the aforementioned distance only requires bit operations, the above distance calculation is quite simple.

The data update and weight update procedures are described below. When an input block is a background block, the weight of the background model is updated by the following equation: w' _k = α M _k +(1 - α ) w _k ; where α is the learning rate, when the best match for the two yuan FIG bit lines M _k is 1, otherwise to 0 M _k lines.

The learning rate determines the speed of adaptation. That is, a larger learning rate has a faster adaptability. Regarding each binary value of the best matching binary bit map, the update rule satisfies the following equation: Where t represents the tth frame, and It is 0 in the initial stage.

If the input block is a foreground block, in the background model, the data update program replaces the binary bit map with the smallest weight value as the input block. In the present invention, the weight value of the new block is set to a lower initial weight value, which is 0.01. Finally, the weight values of the background model are renormalized to a weight value of one.

In the above description, the input block can match a binary bit map with a lower weight value and is considered to be a background block. However, low weight means that the corresponding binary bit map has a lower probability of being a background block. To solve this problem, the present invention sorts the weights of the background model in a decreasing manner, and selects the previous B binary bitmaps as a background model, and the selection method satisfies the following equation: Where TH _B is a predefined threshold.

In addition, when calculating the unit element reference data, the dual-bit reference material, and the three-dimensional reference material, the average number can be generated by the following recursive equation: Where x is a pixel value, R _i is the i-th region and i is greater than 1, M _i is the average of R _i , and |R _i | is the number of pixels in region R _i . Initially, i is equal to 1 and R ₁ is defined as all pixels of a given block.

The K-bit pattern of the pixel can also be generated by repeatedly using the increasing i by the following equation: Referring to Figure 5, there is shown a two-layer recursive procedure for coarse to fine texture descriptions of the present invention. Figure 5(a) is an input block R ₁ having M ₁ of 3.8. Thus, by applying the above-described recursive equation, the block is divided into two sub-blocks R ₂ and R ₃ having M ₂ = 2.17 and M ₃ = 5.76, respectively. This phase will produce the first bit value of the K-bit pattern. Similarly, according to M ₂ and M ₃ , R ₂ will produce R ₄ and R ₅ , and R ₃ will produce R ₆ and R ₇ . Thus, the second bit value of the K bit pattern is generated. It should be noted that if the program ends in octet mode, the most refined level block can be obtained.

It can be observed from the drawing that the size of the area used to evaluate the texture is dynamically determined according to its characteristics, which means that these textures can more accurately represent the block features.

In addition, since each block has a different complexity, each block should be represented as a different bit pattern. We can simplify the variation of the block to determine whether a block is a smooth block, but the calculation of the variance greatly reduces the efficiency of the present invention. Therefore, we use another strategy with relatively low complexity, the number of bitwise transitions of the numbers 1/0 or 0/1 to represent the complexity of a block. The invention is based on 1-bit mode The formula calculates the number of bitwise transitions, and is defined as the sum of the number of bitwise transitions for each column of the binary bit map. For example, the number of bitwise transitions of each column of the binary bit pattern of the 1-bit mode in FIG. 2 is 2, 2, 3, and 2, respectively. If the number of bitwise transitions of the binary bit map is less than a predetermined threshold, the block is a smooth block; otherwise, the block is a complex block.

If the background block is a smooth block, it is represented as a 1-bit mode and the flag value is set to 1, otherwise the non-smooth block is represented as a 2 or 3 bit mode to more accurately fit the block. Characteristic, and its flag value is set to 0. Blocks that the user can define at the outset whether they are not smooth need to be represented in 2-bit or 3-bit mode. In addition, a weight w _s is attached to each background block to indicate whether the block is a smooth block. At the beginning, the w _s system is set to 0.5.

The weight update procedure and the data update procedure are as follows. For an input block, it will use the bitwise transitions to extract its features (smooth or unsmooth) and convert to the corresponding mode, denoted BM _new . Thus, the update of w _s satisfies the following equation: w' _s = α M _s +(1 - α ) w _s ; where, if the input block is a smooth block, M _s is 1.

If the background block is a smooth block, w _s will be close to 1; if the background block is a non-smooth block, w _s is close to 0. The weight and data update procedures are shown in Figure 6. The branch on the right side of Figure 6 is similar to the original 1-bit mode except for the BM _m update procedure, updating the bit pattern The rule is changed from the above equation to the following equation: among them It The system is greater than 1/2 ^k and k is the k-bit mode. For example, in 2-bit mode, if Is "01" and P _ij is {0.4, 0.2, 0.2, 0.2} after the calculation of the above equation is completed, The system changes to "00".

In the left branch of Figure 6, it is used to change the k-bit mode used for the background model. For a smooth background block (flag value 1), when the input block is a non-smooth block and its w _{s is} less than 0.5, the mode of the background block should be changed from 1-bit mode to 2-bit or 3 Bit mode. Therefore, all BM _i using the 1-bit mode are removed, and a 2-bit or 3-bit BM _new is inserted as the first bitmap for the new background model. Finally, let the flag value be 0 and the new background model is reconstructed by using BM _new using 2-bit or 3-bit mode to form K weighted bit maps (BM ₁ , BM ₂ , ... , BM _K ). As for the unsmooth background block (ie, the flag value is 0), when the input block is a smooth block and its ws is greater than or equal to 0.5, the reconstruction step for the new background model is similar to the above description.

Figure 7 is a flow chart showing the instant background modeling method of the present invention for moving body detection. First, BM _new needs to first check if the number of bitwise transitions is greater than the predetermined threshold TH _bt . Based on empirical results, TH _bt can usually be set to 4. After that, the right branch handles whether BM _new is a smooth block, while the left branch indicates a non-smooth block. Finally, both sides have a data update program and a weight update program as shown in FIG.

The experimental results of the use of many databases of the present invention will be shown below and Comparison of methods mentioned in other papers. These databases contain indoor and outdoor scenes. Figures 8 and 9 show indoor scenes with varying brightness, with one participant moving in the direction of the camera. From the results of Figures 8 and 9, it can be observed that the method of Stauffer and Grimson is very sensitive to the change of brightness, while the method of Heikkilä and Pietikäinen is easy. Interference with noise. Conversely, the present invention has a better ability to resist changes in brightness, and because the present invention employs more information, the present invention uses information in one region rather than a single pixel. Since the present invention employs a method of non-overlapping blocks, the method of the present invention calculates a rougher profile than the other methods described above.

A comparison of more outdoor detection results with the methods of Shi Dao Fo and Ge Ningsen is shown in Figures 10 and 11. From the results of Fig. 11, it can be observed that although the present invention may generate more holes in the interior of a moving object, this effect can be eliminated by morphological operations.

In addition, considering the method of the present invention, the comparison and detection results of 1-bit, 2-bit and 3-bit are shown in FIG. From these results, it can be observed that the 2-bit or 3-bit mode has a higher correct rate than the 1-bit mode. Figure 12(c) shows that the 2-bit mode enhances the shape of the moving object. However, in some cases, the shadow may be considered to be foreground information when it is caused by a false positive. In addition, in the 12th (d) figure, more bit combinations are generated, and the improvement of the foreground is more perfect than other results.

Figures 13 to 20 provide the three bit patterns of the present invention. It compares.

It can be observed from Fig. 13 and Fig. 14 that the 1-bit mode and the 2-bit mode have a high probability of destroying the moving object. However, in the 3-bit mode, this damage is corrected and improved. Figure 14 shows that the method of the present invention outperforms other methods in resisting shadows and efficiency.

More comparisons and images using the CAVIAR public database are provided in Figures 15 through 20. The present invention employs a Kewell public database that has been widely used to monitor experimental results of the examples. In Figure 15, a participant is shown walking along the corridor. When the connected component result is displayed in the 2-bit mode of Figure 15(f), the shadow is generated around the foot of the object, and the shadow is considered to be foreground information. However, the 3-bit mode shadow can be appropriately blended into the background.

Please refer to Figure 16, which illustrates the gradual improvement from 1-bit mode to 3-bit mode and the corresponding results of the methods of Storafo and Gingsen and the methods of Hetchira and Petrika. In Figure 16, although the methods of Shi Dao Fo and Ge Ningsen and the methods of Hetchira and Petrika are all possible to obtain a more slender foreground object by using the pixel-based method, the calculation amount is much higher than that based on Block method.

The rest of the experimental results using the Kewell Public Library will use other data commonly used for monitoring examples. In Figures 17 and 18, the method of the present invention can correctly resist the shadow as compared with the methods of Shi Dao Fo and Ge Ningsen. In addition, the hole problem can be mitigated by using a multi-bit mode. However, the methods of Hetchira and Petrika have produced good results in these two problems because they adopt overlapping block strategies, but they also lead Higher calculation costs.

Figure 19 shows that the method of the present invention can detect two separate foreground objects more accurately than other methods. In Fig. 20, although the 1-bit mode blends shadows into the foreground, the 2-bit and 3-bit modes can compete fully with the methods in Hetchira and Petrika.

The performance of the method of the present invention uses a number of image sequences compared to two current state-of-the-art methods. These image sequences are obtained from real indoor and outdoor environments and public test databases. The simulation environment used for this experiment was equipped with a 2.93 GHz Core 2 Duo Intel processor and 2 GB of memory. The image resolution is set at 320x240 pixels. All algorithms are executed in C++.

In order to mark and segment foreground pixels, the connected components algorithm is applied to each background model method. The parameter series used in these experiments is shown in Table 1, where α is the learning rate, TH _B is the threshold used in the equation, K is the Gaussian number, and X means that this method does not require this parameter, BS The method is block size and LBP _{P, R} is only used in the method of Hetchira and Petrika and indicates that the radius R is used to find P neighbors, as shown in Table 1.

The performance comparison of the three methods is shown in Table 2 (the present invention employs a 1-bit mode), with the last column indicating the connected component number (CCL) method used in the background construction. From this table we can observe that the method of the present invention is faster than other methods because the block-based method of the present invention requires only bit operations. The methods in Hetchira and Petrika are the slowest because they divide each frame into overlapping blocks, and the size of the local binary pattern (LBP) histogram significantly affects its performance.

The performance comparison of the three different modes (1-bit, 2-bit, and 3-bit) of the present invention is shown in Table 3. It can be clearly seen that the 1-bit mode has the best efficiency. In addition, Table 4 shows the empirical results of the definition of smooth and non-smooth blocks. As mentioned earlier, two different programs for the variance and bitwise transition plans can be used. The bitwise transformation plan can outperform the variability plan because of its relatively low complexity.

Table 3 shows a comparison of the number of frames displayed per second by the method of the present invention.

In addition, the following experiments were performed in indoor, outdoor, and Kewell public databases. The frames taken by the moving objects are manually labeled to produce ground truth. Simple images contain different conditions, most of which will result in unstable results. Therefore, to analyze this temporary instability, four measurements of precision, recall, similarity, and F-measure are used.

Accuracy can be thought of as an accurate measure, and memory is a measure of completion. The F metric calculates its score in terms of accuracy and memory, which can be interpreted as the harmonic mean of accuracy and memory. The calculations for accuracy, memory, similarity, and F metrics satisfy the following equation: As shown in Tables 5 and 6, the TP, FP, FN, TN, P, and N systems are correctly judged to be positive (True positive), false positives are positive (False positive), false positives are negative (False negative), and correctly judged. It is negative (negative negative), judged to be positive (Total positive), and judged to be negative (Total negative). It should be understood that the above-mentioned scores are between 0 and 1, so higher values represent higher accuracy.

Qualitative assessments can be used to determine whether different methods are competitive. To support qualitative results, we first provide the binary results for the live and multiple methods. Figure 21 to Figure 23 show Show the sequence to be determined. Moreover, once the reality is produced, qualitative results can be obtained intuitively. As described in Tables 7 through 9, the four considered metrics of the present invention can be incrementally increased when more bin patterns are employed, and more accurate results will be obtained.

Moreover, the method of the present invention uses non-overlapping block methods, with problems such as coarse object shapes and hole problems. A more seamless shape can be achieved by implementing a morphological algorithm. Furthermore, as shown in Table 3, the morphological algorithm only slightly affects efficiency. Thus, Figure 24 shows the binary results of the 3-bit mode of the method of the present invention with and without the morphological algorithm. For comparison with pixel-based methods, qualitative measurement systems for indoor, outdoor, and public databases are provided in Tables 10 and 11, respectively.

The above is intended to be illustrative only and not limiting. Any equivalent modifications or alterations to the spirit and scope of the invention are intended to be included in the scope of the appended claims.

S11~S18‧‧‧Step process

R1‧‧‧ input block

R2~R7‧‧‧ Block

Figure 1 is the flow of the instant background modeling method of the present invention Figure.

2 is a schematic diagram of a 1-bit reference material and a 2-bit reference data calculation method of the instant background modeling method of the present invention.

Figure 3 is a schematic diagram showing the calculation method of the 3-bit reference material of the instant background modeling method of the present invention.

Figure 4(a) is an original image of one of the instant background modeling methods of the present invention.

Fig. 4(b) is an image calculated by the texture descriptor according to Fig. 4(a) of the instant background modeling method of the present invention.

Figure 5(a) is an input block R1 which is one of the instant background modeling methods of the present invention.

Figure 5(b) is a block R2 of the instant background modeling method of the present invention.

Figure 5(c) is a block R3 of the instant background modeling method of the present invention.

Figure 5(d) is a 1-bit pattern of the instant background modeling method of the present invention.

Figure 5(e) is a block R4 of the instant background modeling method of the present invention.

Figure 5(f) is a block R5 of the instant background modeling method of the present invention.

The fifth (g) diagram is a block R6 of the instant background modeling method of the present invention.

Figure 5(h) is a block R7 of the instant background modeling method of the present invention.

The 5th (i) diagram is a 2-bit pattern of the instant background modeling method of the present invention.

Figure 6 is a flow chart of the data update procedure and the weight update procedure of the instant background modeling method of the present invention.

Figure 7 is a flow chart of the moving object detection of the instant background modeling method of the present invention.

Figure 8(a) is an original image of one of the instant background modeling methods of the present invention.

Fig. 8(b) is an image calculated by the method of Shi Daofu and Ge Ningsen according to Fig. 8(a) of the instant background modeling method of the present invention.

Fig. 8(c) is an image calculated by the method of Hetchira and Petrika according to Fig. 8(a) of the instant background modeling method of the present invention.

Fig. 8(d) is an image calculated by the 1-bit mode according to Fig. 8(a) of the instant background modeling method of the present invention.

Figure 9(a) is an original image of one of the instant background modeling methods of the present invention.

Figure 9(b) is the basis of the instant background modeling method of the present invention. Figure 9(a) shows the image calculated by the methods of Shi Dao Fo and Ge Ningsen.

Figure 9(c) is an image of the instant background modeling method of the present invention calculated by the method of Hetchira and Petrika according to Fig. 9(a).

Fig. 9(d) is an image calculated by the 1-bit mode according to Fig. 9(a) of the instant background modeling method of the present invention.

Figure 10(a) is an original image of one of the instant background modeling methods of the present invention.

Fig. 10(b) is an image calculated by the method of Shi Daofu and Ge Ningsen according to Fig. 10(a) of the instant background modeling method of the present invention.

Fig. 10(c) is an image calculated by the 1-bit mode according to Fig. 10(a) of the instant background modeling method of the present invention.

Figure 11(a) is an original image of one of the instant background modeling methods of the present invention.

Fig. 11(b) is an image calculated by the method of Shi Dao Fo and Ge Ningsen according to the image of Fig. 11(a) of the instant background modeling method of the present invention.

Fig. 11(c) is an image calculated by the instant background modeling method of the present invention according to Fig. 11(a).

Fig. 11(d) is an image calculated by the instant background modeling method of the present invention according to Fig. 11(a) and the morphological algorithm.

Figure 12(a) is an original image of one of the instant background modeling methods of the present invention.

Fig. 12(b) is an image calculated by the 1-bit mode according to Fig. 12(a) of the instant background modeling method of the present invention.

Fig. 12(c) is an image calculated by the 1-bit and 2-bit modes according to Fig. 12(a) of the instant background modeling method of the present invention.

Fig. 12(d) is an image calculated by the 1-bit and 3-bit modes according to Fig. 12(a) of the instant background modeling method of the present invention.

Fig. 12(e) is an image of a connected component result (CCL) calculated in a 1-bit mode according to Fig. 12(a) of the instant background modeling method of the present invention.

Fig. 12(f) is an image of the connected component result calculated by the 1-bit and 2-bit modes according to Fig. 12(a) of the instant background modeling method of the present invention.

Fig. 12(g) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 12(a) of the instant background modeling method of the present invention.

Figure 13(a) is an original image of one of the instant background modeling methods of the present invention.

Figure 13(b) is the basis of the instant background modeling method of the present invention. Figure 13(a) shows the image calculated in 1-bit mode.

Fig. 13(c) is an image calculated by the 1-bit and 2-bit modes according to Fig. 13(a) of the instant background modeling method of the present invention.

Fig. 13(d) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 13(a) of the instant background modeling method of the present invention.

Fig. 13(e) is an image of the connected component result calculated in the 1-bit mode according to Fig. 13(a) of the instant background modeling method of the present invention.

Fig. 13(f) is an image of the connected component result calculated by the 1-bit and 2-bit modes according to Fig. 13(a) of the instant background modeling method of the present invention.

Fig. 13(g) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 13(a) of the instant background modeling method of the present invention.

Figure 14(a) is an original image of one of the instant background modeling methods of the present invention.

Fig. 14(b) is an image calculated by the 1-bit mode according to Fig. 14(a) of the instant background modeling method of the present invention.

Fig. 14(c) is an image calculated by the 1-bit and 2-bit modes according to Fig. 14(a) of the instant background modeling method of the present invention.

Fig. 14(d) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 14(a) of the instant background modeling method of the present invention.

Fig. 14(e) is an image of the connected component result calculated in the 1-bit mode according to Fig. 14(a) of the instant background modeling method of the present invention.

Fig. 14(f) is an image of the connected component result calculated by the 1-bit and 2-bit modes according to Fig. 14(a) of the instant background modeling method of the present invention.

Fig. 14(g) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 14(a) of the instant background modeling method of the present invention.

Fig. 14(h) is an image calculated by the method of Shi Daofu and Ge Ningsen according to Fig. 14(a) of the instant background modeling method of the present invention.

Figure 14(i) is an image of the instant background modeling method of the present invention calculated according to the method of Haciilla and Petrika according to Fig. 14(a).

Figure 15(a) is an original image of one of the instant background modeling methods of the present invention.

Fig. 15(b) is an image calculated by the 1-bit mode according to Fig. 15(a) of the instant background modeling method of the present invention.

Figure 15(c) is the basis of the instant background modeling method of the present invention. Figure 15(a) shows the image calculated in 1-bit and 2-bit modes.

Fig. 15(d) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 15(a) of the instant background modeling method of the present invention.

Fig. 15(e) is an image of the connected component result calculated in the 1-bit mode according to Fig. 15(a) of the instant background modeling method of the present invention.

Fig. 15(f) is an image of the connected component result calculated by the 1-bit and 2-bit modes according to Fig. 15(a) of the instant background modeling method of the present invention.

Fig. 15(g) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 15(a) of the instant background modeling method of the present invention.

Figure 16(a) is an original image of one of the instant background modeling methods of the present invention.

Fig. 16(b) is an image calculated by the 1-bit mode according to Fig. 16(a) of the instant background modeling method of the present invention.

Fig. 16(c) is an image calculated by the 1-bit and 2-bit modes according to Fig. 16(a) of the instant background modeling method of the present invention.

Figure 16(d) is a calculation of the instant background modeling method of the present invention based on the 16-bit (a) and 1-bit modes. The image of the connected component result.

Fig. 16(e) is an image of the connected component result calculated in the 1-bit mode according to Fig. 16(a) of the instant background modeling method of the present invention.

Fig. 16(f) is an image of the connected component result calculated by the 1-bit and 2-bit modes according to Fig. 16(a) of the instant background modeling method of the present invention.

Fig. 16(g) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 16(a) of the instant background modeling method of the present invention.

Fig. 16(h) is an image calculated by the method of Stora and Gingsen according to Fig. 16(a) of the instant background modeling method of the present invention.

Fig. 16(i) is an image of the instant background modeling method of the present invention calculated according to the method in Fig. 16(a) in Haciilla and Petrika.

Figure 17(a) is an original image of one of the instant background modeling methods of the present invention.

Fig. 17(b) is an image calculated by the 1-bit mode according to Fig. 17(a) of the instant background modeling method of the present invention.

Fig. 17(c) is an image calculated by the 1-bit and 2-bit modes according to Fig. 17(a) of the instant background modeling method of the present invention.

Fig. 17(d) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 17(a) of the instant background modeling method of the present invention.

Fig. 17(e) is an image of the connected component result calculated in the 1-bit mode according to Fig. 17(a) of the instant background modeling method of the present invention.

Fig. 17(f) is an image of the connected component result calculated by the 1-bit and 2-bit modes according to Fig. 17(a) of the instant background modeling method of the present invention.

Fig. 17(g) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 17(a) of the instant background modeling method of the present invention.

Figure 17(h) is an image of the instant background modeling method of the present invention calculated according to the method of Stora and Gingsen according to Fig. 17(a).

Figure 17(i) is an image of the instant background modeling method of the present invention calculated according to the method of Haciilla and Petrika according to Fig. 17(a).

Figure 18(a) is an original image of one of the instant background modeling methods of the present invention.

Fig. 18(b) is an image calculated by the 1-bit mode according to Fig. 18(a) of the instant background modeling method of the present invention.

Figure 18(c) is the basis of the instant background modeling method of the present invention. Figure 18(a) shows the image calculated in 1-bit and 2-bit modes.

Fig. 18(d) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 18(a) of the instant background modeling method of the present invention.

Fig. 18(e) is an image of the connected component result calculated in the 1-bit mode according to Fig. 18(a) of the instant background modeling method of the present invention.

Fig. 18(f) is an image of the connected component result calculated by the 1-bit and 2-bit modes according to Fig. 18(a) of the instant background modeling method of the present invention.

Fig. 18(g) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 18(a) of the instant background modeling method of the present invention.

Figure 18(h) is an image of the instant background modeling method of the present invention calculated according to the method of Stora and Gingsen according to Fig. 18(a).

Figure 18(i) is an image of the instant background modeling method of the present invention calculated according to the method of Haciilla and Petrika according to Fig. 18(a).

Figure 19(a) is an original image of one of the instant background modeling methods of the present invention.

Figure 19(b) is the basis of the instant background modeling method of the present invention. Figure 19(a) shows the image calculated in 1-bit mode.

Fig. 19(c) is an image calculated by the 1-bit and 2-bit modes according to Fig. 19(a) of the instant background modeling method of the present invention.

Fig. 19(d) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 19(a) of the instant background modeling method of the present invention.

Fig. 19(e) is an image of the connected component result calculated in the 1-bit mode according to Fig. 19(a) of the instant background modeling method of the present invention.

Fig. 19(f) is an image of the connected component result calculated by the 1-bit and 2-bit modes according to Fig. 19(a) of the instant background modeling method of the present invention.

Fig. 19(g) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 19(a) of the instant background modeling method of the present invention.

Fig. 19(h) is an image calculated by the method of Shi Daofu and Ge Ningsen according to the 19th (a) diagram of the instant background modeling method of the present invention.

Figure 19(i) is an image of the instant background modeling method of the present invention calculated according to the method of Haciilla and Petrika according to Fig. 19(a).

Figure 20(a) is one of the instant background modeling methods of the present invention. Start image.

Fig. 20(b) is an image calculated by the one-bit mode according to Fig. 20(a) of the instant background modeling method of the present invention.

Fig. 20(c) is an image calculated by the 1-bit and 2-bit modes according to Fig. 20(a) of the instant background modeling method of the present invention.

Fig. 20(d) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 20(a) of the instant background modeling method of the present invention.

Fig. 20(e) is an image of the connected component result calculated in the 1-bit mode according to Fig. 20(a) of the instant background modeling method of the present invention.

Fig. 20(f) is an image of the connected component result calculated by the 1-bit and 2-bit modes according to Fig. 20(a) of the instant background modeling method of the present invention.

Fig. 20(g) is an image of the connected component result calculated by the 1-bit and 3-bit modes according to Fig. 20(a) of the instant background modeling method of the present invention.

Fig. 20(h) is an image calculated by the method of Shi Daofu and Ge Ningsen according to Fig. 20(a) of the instant background modeling method of the present invention.

Figure 20(i) is a method for instant background modeling of the present invention based on the method of Hochila and Petrika according to Figure 20(a) Calculated image.

Figure 21(a) is an image of the ground truth of the instant background modeling method of the present invention.

Figure 21(b) is a binary image calculated according to the 21st (a) figure of the instant background modeling method of the present invention.

Fig. 21(c) is a binary image calculated by the instant background modeling method of the present invention in a 2-bit mode according to Fig. 21(a).

Fig. 21(d) is a binary image calculated by the instant background modeling method of the present invention in a 3-bit mode according to Fig. 21(a).

Figure 22(a) is an image of the live state of the instant background modeling method of the present invention.

Fig. 22(b) is a binary image calculated according to Fig. 22(a) of the instant background modeling method of the present invention.

Fig. 22(c) is a binary image calculated by the instant background modeling method of the present invention in a 2-bit mode according to Fig. 22(a).

Fig. 22(d) is a binary image calculated by the instant background modeling method of the present invention in a 3-bit mode according to Fig. 22(a).

Figure 23(a) is an image of the live state of the instant background modeling method of the present invention.

Fig. 23(b) is a binary image calculated according to Fig. 23(a) of the instant background modeling method of the present invention.

Fig. 23(c) is a binary image calculated by the instant background modeling method of the present invention in a 2-bit mode according to Fig. 23(a).

Fig. 23(d) is a binary image calculated by the instant background modeling method of the present invention in a 3-bit mode according to Fig. 23(a).

Fig. 24(a) is an image of the instant background modeling method of the present invention calculated by a 3-bit pattern and a morphological algorithm based on an image of an indoor sequence.

Fig. 24(b) is an image obtained by the instant background modeling method of the present invention based on the image of the second indoor sequence in a 3-bit mode and a morphological algorithm.

Fig. 24(c) is an image of the instant background modeling method of the present invention calculated by a 3-bit pattern and a morphological algorithm according to the image of the Caviery (CAVIAR) WalkByShop1front database.

Figure 24(d) is an image of the instant background modeling method of the present invention calculated by a 3-bit pattern and a morphological algorithm based on images of the Kavier's ShopAssistant1cor database.

Figure 24(e) is an image of the instant background modeling method of the present invention calculated by a 3-bit pattern and a morphological algorithm based on the image of the OneStopMoveNoEnter2cor database of Kewell.

S11~S18‧‧‧Step process

Claims (9)

An instant background modeling method is applicable to a surveillance camera device, wherein the surveillance camera device is provided with an image capture module and a processing module, and the method comprises the following steps: capturing the plurality of images by the image capture module Each of the reference images is divided into a plurality of reference blocks that do not overlap each other through the processing module; and the processing module calculates a unit cell reference according to each of the reference blocks of each of the reference images. Data and a double-bit reference data; the image capturing module captures a real-time image; the processing module divides the real-time image into a plurality of non-overlapping instant blocks according to the cutting manner of the reference image Using the processing module to calculate a unit of real-time data and a pair of real-time data according to each of the real-time blocks; using the processing module to perform the unit-yuan real-time data and the corresponding unit-element reference data Comparing, or comparing the two-bit real-time data with each of the two-bit reference materials corresponding to the position to respectively generate a matching result; The processing module selectively executes a data update procedure according to each of the matching results to update each of the unit cell reference materials or each of the dual-bit reference materials; wherein each of the unit cell reference materials and the unit cell real-time data meets the following conditions : Wherein, TH _smooth is a preset threshold, x _ij is the pixel value of each reference block or the position (i, j) of the instant block, and m is each of the reference block or the instant block. The average value, a _ij is the unit value of one of the unit element reference materials or the location (i, j) of the unitary metadata. The method for instant background modeling according to claim 1, further comprising the steps of: calculating, by the processing module, the number of bit transitions according to each of the unit reference data of each of the reference images; and The processing module compares each of the bit transition times with a smoothing threshold, and if the number of bit transitions is less than or equal to the smoothing threshold, the processing module selectively selects the unit cell instantaneously The data is compared with each of the unit cell reference data corresponding to the location, and if the number of bit transitions is greater than the smoothing threshold, the processing module selectively utilizes the dual-bit real-time data and the location corresponding to each Double-bit reference data for comparison. The method for instant background modeling according to claim 1 further includes the following steps: calculating, by the processing module, a unit distance according to the unit element reference data corresponding to the unit element real-time data and the location Value, or the double-bit real-time data corresponding to the location The bit reference data respectively calculates a double bit distance value; and the processing module compares each of the unit cell distance values or each of the double bit distance values with a distance threshold to generate the matching result; And if the minimum value of each unit cell distance value or the minimum value of each of the double bit distance values is less than the distance threshold, the processing module executes a weight update procedure, and each of the matching results When the unit distance value or each of the double bit distance values is greater than the distance threshold, the data update procedure is executed by the processing module. The instant background modeling method according to claim 3, wherein the weight update program comprises the following steps: updating, by the processing module, one unit weight value of each unit reference material or each of the double digits One of the meta-references is a double-bit weight value. The instant background modeling method as described in claim 1, wherein the data update program further comprises the following steps: sorting, by using the processing module, a unit weight value of each unit reference material, or Each of the two-bit reference materials is sorted by a double-bit weight value; the unit element reference material having the smallest value among the unit weight values is replaced by the processing unit to the unit-yuan real-time data, or each The double-bit reference material having the smallest value among the double-bit weight values is replaced by the dual-bit real-time data; The processing module replaces the unit weight value of the minimum value or the double value weight value of the minimum value with an initial weight value. The instant background modeling method as described in claim 1, wherein each of the dual-bit reference materials and the dual-bit real-time data satisfy the following conditions: Wherein, TH _smooth is a preset threshold, x _ij is the pixel value of each reference block or the position (i, j) of the instant block, and m is each of the reference block or the instant block. The average value, hm is the average value calculated when the unit value of the unit reference material of each reference block is 1 or the unit value of the real-time data of the real-time block is 1 The average calculated by the time is lm is the average calculated separately when the unit value of the unit reference data of each reference block is 0 or the unit of the real-time data of the real-time block The average calculated by the value of 0 is b _ij is the double-bit value of each of the two-bit reference data or the position (i, j) of the dual-bit real-time data. The instant background modeling method as described in claim 1 further includes the following steps: by the processing module, according to each of the unit reference materials and each The two-bit reference data is used to calculate a three-digit reference data, and one-third real-time data is calculated based on the real-time data of the unit and the real-time data of the two-digit real-time data; and the three-dimensional instant using the processing module The data is compared with each of the three-bit reference materials corresponding to the location to respectively generate the matching result, and each of the three-bit reference materials is selectively updated according to each matching result. For example, the instant background modeling method described in claim 7 wherein each of the three-dimensional reference material and the three-dimensional instant data satisfy the following conditions: Wherein, TH _smooth is a preset threshold, x _ij is the pixel value of each reference block or the position (i, j) of the instant block, and m is each of the reference block or the instant block. The average value, hm is the average value calculated when the unit value of the unit reference material of each reference block is 1 or the unit value of the real-time data of the real-time block is 1 The average calculated by the time is lm is the average calculated separately when the unit value of the unit reference data of each reference block is 0 or the unit of the real-time data of the real-time block The average number calculated when the value is 0, hhm is the average calculated separately when the double-bit value of the double-bit reference data of each reference block is 11, or the real-time block The average of the double-bit real-time data when the double-bit value is 11, and the hlm is the average calculated when the double-bit value of the double-bit reference data of each reference block is 10. The number or the average of the double-bit real-time data of the instant block when the double-bit value is 10, the lhm is The average of the two-bit reference data of the reference block when the double-bit value is 01 or the double-bit value of the dual-bit real-time data of the real-time block is calculated as 01. The average number of llm is the average of the two-bit real-time data of the real-time block when the double-bit value of the double-bit reference data of each reference block is 00 The average value calculated when the value is 00, and c _ij is one of the three-bit values of the three-bit reference material or the position (i, j) of the three-dimensional real-time data. The instant background modeling method as described in claim 7 further includes the following steps: when the processing module compares the three-dimensional real-time data with the three-dimensional reference data corresponding to the location, Calculating a three-bit distance value separately, and comparing each of the three-bit distance values with a distance threshold to generate the matching result; If the minimum value of each of the three-bit distance values is less than the distance threshold, the processing module executes a weight update procedure, and each of the three-bit distance values is greater than the distance threshold. And the data update program is executed by the processing module.