CN105335782A - Image-based target object counting method and apparatus - Google Patents

Abstract

The present invention provides an image-based target object counting method and an image-based target object counting device, wherein the image-based target object counting method includes: extracting a plurality of sub-image blocks from the current image, each sub-image block It is an area outside the boundary line of the predetermined statistical area and at the same time bounded by the boundary line; for each sub-image block, the foreground image of the sub-image block is obtained, and the fusion target is obtained from the foreground image, and each fusion target represents a passing target object, compare the fusion target obtained from the sub-image block in the current image and at least the same sub-image block in the previous image with the boundary line, so as to perform the counting of the number of target objects; perform the statistical results of multiple sub-image blocks Fusion, the number of target objects entering the predetermined statistical area is obtained. The invention can count the number of target objects entering a specific area based on images, and can solve the problem of repeated counting.

Description

Image-based target object counting method and device

technical field

The present invention relates to the technical field of image processing, in particular to an image-based target object counting method and an image-based target object counting device.

Background technique

Real-time people flow information is a very useful resource for people management and safety applications such as pedestrian flow management, passenger and passenger flow estimation. For example, statistics obtained by counting people in a shopping mall can be used to help businesses convert potential visitors into customers to enhance customer experience, thereby effectively improving business performance. In addition, counting passengers in public traffic vehicles can help improve the management of public traffic vehicles and increase efficiency.

In particular, if the flow of people in a certain area can be obtained, for example, the area in front of a glass display case can be used to understand the degree of people's interest in a corresponding product or location. People flow information with time information can be further analyzed and improved.

Contents of the invention

In view of this, the present invention proposes a new image-based target object counting technology to at least solve the problem of how to obtain crowd flow information in a certain area.

In view of this, according to one aspect of the present invention, an image-based target object counting method is provided, including: extracting a plurality of sub-image blocks from the current image, each of which is outside a boundary line of a predetermined statistical area And at the same time, the region bordered by the boundary line; for each sub-image block, obtain the foreground image of the sub-image block, and obtain the fusion target from the foreground image, each fusion target represents a passing target object, comparing the fusion target obtained from the sub-image block in the current image and at least the same sub-image block in the previous image with the boundary line, so as to perform the counting of the number of target objects; The statistical results are fused to obtain the number of target objects entering the predetermined statistical area.

According to another aspect of the present invention, there is also provided an image-based target object counting device, including: an extraction unit that extracts a plurality of sub-image blocks from the current image, and each of the sub-image blocks is a boundary line of a predetermined statistical area The area outside and at the same time bordered by the boundary line; the fusion target acquisition unit, for each sub-image block, obtains the foreground image of the sub-image block, and obtains the fusion target from the foreground image, and each fusion target Representing a passing target object; the counting unit compares the fusion target obtained from the sub-image block in the current image and the same sub-image block in at least the previous image with the boundary line to perform target object counting; The statistical result fusion unit is configured to fuse the statistical results of the multiple sub-image blocks to obtain the number of target objects entering the predetermined statistical area.

According to still another aspect of the present invention, there is also provided an electronic device, the electronic device comprising the above-mentioned apparatus for counting target objects based on an image.

According to still another aspect of the present invention, there is also provided a program product storing machine-readable instruction codes. When the program product is executed, the above-mentioned machine can execute the image-based target object counting method as described above.

In addition, according to other aspects of the present invention, there is also provided a computer-readable storage medium on which the above-mentioned program product is stored.

The image-based target object counting device, the image-based target object counting method, and the electronic device according to the above-mentioned embodiments of the present invention divide the image outside the statistical area into a plurality of sub-image blocks, count the target objects of each sub-image block separately, and The fusion processing of the target objects of multiple sub-image blocks can achieve at least one of the following beneficial effects: the number of target objects entering the statistical area can be obtained; a fusion algorithm is provided to prevent repeated counting.

These and other advantages of the present invention will be more apparent through the following detailed description of the preferred embodiments of the present invention with reference to the accompanying drawings.

Description of drawings

The present invention can be better understood by referring to the following description given in conjunction with the accompanying drawings, wherein the same or similar reference numerals are used throughout to designate the same or similar parts. The accompanying drawings, together with the following detailed description, are incorporated in and form a part of this specification, and serve to further illustrate preferred embodiments of the invention and explain the principles and advantages of the invention. In the attached picture:

FIG. 1 shows a schematic diagram of an image-based target object counting method according to an embodiment of the present invention;

Fig. 2 shows a schematic diagram of defining a statistical area according to another embodiment of the present invention;

Fig. 3 shows a schematic diagram of an image-based target object counting method according to another embodiment of the present invention;

FIG. 4 shows a schematic diagram of extracted sub-image blocks according to an embodiment of the present invention;

Fig. 5 shows a schematic diagram of people counting according to an embodiment of the present invention;

Fig. 6 shows a schematic diagram of an image-based target object counting method according to yet another embodiment of the present invention;

Figure 7 shows an example of merging candidate targets into a fusion target, where Figure 7(a) represents a situation where two candidate targets overlap each other, and Figure 7(b) represents a situation where the distance between candidate targets is less than a predetermined distance;

Fig. 8 shows a schematic diagram of boundaries and corners of statistical regions according to an embodiment of the present invention;

Fig. 9 shows a schematic diagram of a sub-image block containing a target object under the scene shown in Fig. 8;

FIG. 10 shows a schematic diagram of the distribution of the information of each target object on the time axis;

Fig. 11 shows a block diagram of an image-based target object counting device according to an embodiment of the present invention.

detailed description

In order to understand the above-mentioned purpose, features and advantages of the present invention more clearly, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the case of no conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

In the following description, many specific details are set forth in order to fully understand the present invention, but the present invention can also be implemented in other ways different from those described here, therefore, the present invention is not limited to the specific embodiments disclosed below limit.

Fig. 1 shows a schematic diagram of an image-based target object counting method according to an embodiment of the present invention.

As shown in Figure 1, the image-based target object counting method according to an embodiment of the present invention may include the following steps:

Step 102, receiving an input video image, and setting a predetermined statistical area on the image. In this step, a statistical area is defined, and the number of people entering the statistical area is counted.

In step 104, each boundary of the predetermined area may be called a counting line, and the number of people crossing the counting line and entering the counting area is counted.

In step 106, the statistical combination of all counting lines is merged, so as to prevent the same person in the same scene from being counted by different counting lines.

In the above step 102, as shown in FIG. 2, the boundary line of the statistics area and the corresponding statistics direction are set on the image.

According to Figure 2, it can be seen that the statistical area is a closed polygon, which includes several counting lines, and each counting line can be seen as the boundary of the statistical area. In this embodiment, the statistical area is a quadrilateral, including four counting lines. Each count line has a statistical method that defines the direction into that statistical region. If a person crosses the count line from outside and enters the count area, a count will be performed.

Above-mentioned step 104 can comprise two sub-steps, as shown in Figure 3:

Step 302, extract multiple sub-image blocks from the current image. Wherein, the sub-image block is an area outside a boundary line of the predetermined statistical area and at the same time bounded by the boundary line, see FIG. 4 .

Step 304, after the sub-image blocks are extracted, count the number of people based on each sub-image block. Step 304 may further include three parts: adaptive foreground segmentation, obtaining fused objects and counting fused objects.

In adaptive foreground segmentation, the foreground image containing motion area information in the current sub-image block is obtained through the difference between the current sub-image block and the reference sub-image block, and then the foreground image is filtered, and the foreground image is a binary image , such as the foreground image of the sub-image block in Figure 5.

Then, fused objects are obtained from the foreground image, each fused object represents a passing person. Specifically, once the foreground image is obtained, moving objects are segmented from the foreground image, and each moving object is used as a candidate for a passing person. If the number of candidate targets is greater than 0, multiple candidate targets are fused according to preset rules to obtain a fused target, see FIG. 5 .

Finally, people counting is performed based on a comparison of the fusion target obtained from the current image and at least the previous image with preset statistical boundary lines. If the number of fusion targets is zero, skip this step, otherwise execute this step.

FIG. 6 shows a flow chart of specific processing of the image-based people counting method according to the present invention. As shown in FIG. 6, the foreground image acquisition part includes steps 602-606. Specifically, first, in step 602, for each sub-image block in the current image (for the definition of the sub-image block, refer to the definition above, which will not be repeated here), the difference image between the sub-image block and the reference image is calculated. The reference image is an image block of the same area in an image captured before the current image, and in an initial state, an average environmental background image may be used as a reference image. In the calculation of the difference image, when the input image is an RGB image, the three components of red (R), green (G) and blue (B) that constitute each pixel are respectively calculated between the current image and the reference image. The absolute difference is then averaged to obtain the pixel value of the corresponding pixel in the difference image, and thus the difference image is obtained. Of course, for a grayscale image, the black, white, and gray components can also be similarly used to obtain a differential image. The resulting differential image is a grayscale image.

Next, in step 604, filtering is performed on the difference image to obtain a filtered image. In one embodiment, the present invention employs a median filtering technique. Specifically, for a certain pixel, the median value is found from the values of all surrounding pixels of the pixel, and the median value is used as the value of the pixel. In this way, a filtered image is obtained. Although the median filtering technique is utilized in this description, other filtering methods commonly used in the art may also be utilized.

Next, in step 606, the foreground image is obtained by assigning first values to pixels whose pixel values are greater than or equal to an adaptive segmentation threshold in the filtered image and assigning second values to other pixels. Specifically, if the value of a pixel in the differential image is greater than or equal to the adaptive segmentation threshold, the pixel value is assigned 1 (that is, representing a white point) as a foreground pixel, otherwise it is assigned 0 (that is, representing a black point, background pixels). The values here are only exemplary, and other values can also be used as long as they distinguish the foreground pixels from the background pixels. It can be seen that the foreground image is a binary image.

Through the above steps 602-606, the foreground image composed of the information of the motion area is obtained from the input current image. That is, the influence of the environmental background is removed. It should be noted that an adaptive segmentation threshold is used in the present invention, wherein the adaptive segmentation threshold is the product obtained by multiplying the average value of pixels in the filtered image by a predetermined coefficient and the predetermined minimum segmentation threshold. the big one. The predetermined coefficient and the predetermined minimum segmentation threshold used here are set according to experience to remove the influence of the environmental background as much as possible. If the predetermined coefficient is set too small, the predetermined minimum segmentation threshold is used as the adaptive segmentation threshold, so that there are more white spots in the obtained foreground image, and more fusion objects will be obtained in the fusion object acquisition described later. Conversely, if the predetermined coefficient is set too large, fewer fusion objects will be obtained. Too many fusion targets will increase the amount of calculations for subsequent processing, while too few fusion targets will affect the accuracy of people counting. Therefore, those skilled in the art can set appropriate values based on experience to obtain desired results.

After obtaining the foreground image as described above, the method proceeds to the fused object obtaining part. Continuing to refer to FIG. 6 , the fused target acquisition part includes steps 608 - 612 . Specifically, in step 608, the smallest rectangle that includes the connected region of pixels with the first value in the foreground image and satisfies the first preset condition is used as a candidate target, and each candidate target represents a passing person, where , the first preset condition is that the number of pixels with the first value in the smallest rectangle serving as the candidate target is not less than a predetermined number. In the example above, the pixels assigned the value "1" are regarded as foreground pixels, that is, represented as white dots, therefore, the smallest rectangle containing a connected region of white dots and wherein the number of white dots is not less than a predetermined number is regarded as a candidate object .

Next, as shown in FIG. 6 , it is judged whether to obtain a candidate target (step 610 ). In the case that no candidate target is obtained (step 610, "No"), it is determined that there is no person in the current image, so there is no need to perform subsequent steps, so the method proceeds to step 620 to end the processing of the current image. On the other hand, in case a candidate target is obtained (step 610, YES), the method proceeds to step 612 to obtain a fused target representing a passing person.

Specifically, in step 612, fusion is performed on the obtained candidate targets according to predetermined rules to obtain the fusion target. The fusion mentioned here refers to taking the set of candidate targets that meet the requirements as a collection, wherein the predetermined rule includes: the smallest rectangle that includes the candidate targets that overlap each other among the obtained candidate targets is used as the fusion target, as shown in Figure 7 ( Shown in a); Include the minimum rectangle of the candidate target whose distance between each other in the obtained candidate target is not greater than the predetermined distance as the fusion target, as shown in Fig. 7 (b); and the obtained candidate target Candidate targets that do not overlap with any other candidate targets and have a distance greater than the predetermined distance from any other candidate targets are individually used as the fusion target, as shown in FIG. 7( b ). In addition to the above limitations, the area of the fusion target should not be smaller than a predetermined area. This is because, if the area of the fused object is too small, the fused object is unlikely to represent a person, so such an area will be ignored, thereby reducing the amount of computation.

In the case that the fusion target is not obtained, it is determined that there is no statistical object in the current frame image, so the following steps do not need to be performed, so the method proceeds to step 620 to end the processing of the current frame image. On the other hand, in case a candidate target is obtained, the method proceeds to the people counting part.

Next, as shown in FIG. 6 , it is judged whether to obtain a fusion target (step 614 ). In the case of not obtaining fusion (step 614, "No"), it is determined that there is no person in the current image, so there is no need to perform subsequent steps, so the method proceeds to step 620 to end the processing of the current image. On the other hand, in the case that the fusion target is obtained (step 614, "Yes"), the method proceeds to step 616 for subsequent processing.

As shown in FIG. 6, the people counting part includes steps 616-620. Specifically, in step 616, a fusion target satisfying a second preset condition is found from the obtained fusion targets as the object fusion target, wherein the second preset condition is that the fusion target as the object fusion target has The ratio of the area occupied by the pixels of the first value (ie, foreground pixels) to the entire area occupied by the fused object is greater than a predetermined ratio threshold. If no such object fusion target is found (step 616, "No"), the method proceeds to step S411 to end the process.

On the other hand, when there is such an object fusion target found (step 616, “Yes”), the method proceeds to step 618 .

Specifically, in step 618, it is judged whether the object fusion target in the sub-image block of the current image exceeds the counting line, and if so, for the object fusion target in the sub-image block of the current image, the The fusion object having the largest overlapping area with the object fusion object and the largest overlapping area exceeding a predetermined area threshold is found among the fusion objects as the most overlapping fusion object of the object fusion object. That is, the past location of the person identified in the current image is found from previous images.

Afterwards, comparing each of the object fusion targets in the current image and at least the corresponding most overlapping fusion target in the previous image with a preset statistical boundary line to determine whether to increment the statistical value, wherein, if the object fusion in the current image The statistic value is incremented when an object crosses a predetermined statistical boundary line and the corresponding most overlapping fused object in at least the previous image does not cross the same predetermined statistical boundary line. That is to say, the person of interest in two consecutive frames of images has crossed the preset statistical boundary, that is, the person is passing, so the statistical value is incremented.

As mentioned above, comparing the position of the person detected in the current image and the position of the person in the previous image with the preset statistical boundary line can determine whether a person has passed. It should be noted that the above processing is performed on each sub-image block, so as to obtain the result of counting people corresponding to each sub-image block.

In addition, it should be noted that when there is no fusion target that overlaps most with the fusion target of the current image in the previous frame image, the present invention can find out the fusion target that overlaps the most with the fusion target of the current image from the previous second frame image and The overlapping area meets the required fusion target, so it is determined whether to count by comparing with the preset boundary line. Such a situation is shown in FIG. 6 . By setting in this way, in the case that the most overlapping fusion target is not found in the previous frame image, the most overlapping target may be found in the previous second frame image, thus improving the statistical accuracy. Of course, there is a situation that neither the previous frame image nor the previous second frame image can find the most overlapped fusion target. At this time, counting is not performed.

After the people counting of the current image, the method of the present invention proceeds to step 620 to update the relevant parameters and the reference image so as to process the next frame of image.

Next, with reference to FIG. 8 to FIG. 10 , the processing process of statistically combining multiple sub-image blocks according to the present invention will be described in detail.

When performing target object statistics according to the sub-image blocks, record the information of the target objects that cross the boundary line and enter the predetermined statistical area. For example, if a person crosses boundary line n from position P and enters a predetermined statistical area at time t, the recorded information is (n, t, p). If the predetermined statistical area has N corners (the angle between two adjacent boundary lines), then the value of p is in the range of 1 to N. In this embodiment, the predetermined statistical area has four corners, and FIG. 8 shows the positions of these corners. If the value of p is 4, it means that the target object has passed through corner 4, and if the value of p is 0, it means that the target object has not passed through any corners.

Assuming that FIG. 8 is a current frame image, multiple sub-image blocks can be extracted from the current frame image shown in FIG. 8 , and FIG. 9 shows three sub-image blocks with a target object. Target object I enters the predetermined statistical area from corner 4 and is counted by boundary lines 4 and 3 at the same time; target object II enters the predetermined statistical area from corner 1 and is counted by boundary lines 4 and 1 at the same time; target object III enters the predetermined statistical area through boundary line 2 The statistical area is counted by the boundary line 2. Therefore, the target object entering the predetermined statistical area from the corner will be counted repeatedly, and the present invention proposes a processing method to solve the problem of repeated counting:

If it is determined that the recorded passing positions (that is, the corners passed) of the two target objects are the same, and the difference between the statistical time points of the two target objects is less than the preset time period, then determine the two target objects The object is the same target object that enters the predetermined statistical area at the same moment, and the count of the target object whose statistical time point is later among the two target objects is ignored.

As shown in Figure 10, since the target object (1, t, 1) and the target object (2, t+k, 1) pass through the same corner 1, and k is less than the preset time period, the target object (1, t, 1) The target object (2, t+k, 1) is the same target object that entered the predetermined statistical area at the same time, that is, the target object (2, t+k, 1) is repeatedly counted, in order to facilitate subsequent statistics Calculate and discard the target object (2, t+k, 1), that is, the repeated target object after the time point.

Fig. 11 shows a block diagram of a target object counting device according to an embodiment of the present invention.

As shown in FIG. 11 , an image-based target object counting device 1100 according to an embodiment of the present invention may include:

Extracting unit 1102, extracting a plurality of sub-image blocks from the current image, each sub-image block is an area outside a boundary line of a predetermined statistical area and at the same time bounded by the boundary line;

The fusion target acquisition unit 1104, for each sub-image block, obtains the foreground image of the sub-image block, and obtains the fusion target from the foreground image, and each fusion target represents a passing target object;

A counting unit 1106, which compares the fusion target obtained from the sub-image block in the current image and at least the same sub-image block in the previous image with the boundary line, so as to perform counting of the number of target objects;

The statistical result fusion unit 1108 fuses multiple statistical results based on sub-image blocks to obtain the number of target objects entering a predetermined statistical area.

Wherein, the counting unit 1106 includes: a recording unit 1106A, which records the information of each target object that crosses the boundary line and enters a predetermined statistical area when performing target object statistics according to sub-image blocks.

The statistical result fusion unit 1108 includes: a judging unit 1108A, which judges whether the same target object entering the predetermined statistical area at the same time is counted repeatedly according to the information of the target object; and a screening unit 1108B, which screens the statistical results according to the judgment result.

Wherein, the information includes the statistical time point, the passing position, and the passing position is the angle position where two boundary lines of the predetermined statistical area intersect; When the difference between the statistical time points of two target objects is less than the preset time period, it is determined that the two target objects are the same target object that enters the predetermined statistical area at the same time; the screening unit 1108B ignores that the statistical time point of the two target objects is later The count of target objects for .

Wherein, the fusion target acquisition unit 1104 includes:

The candidate target determining unit 1104A, takes the connected area of pixels with preset values in the foreground image and the smallest rectangle that satisfies the first preset condition as candidate targets, each candidate target represents a candidate for a passing target object, wherein, the first A preset condition is that the number of pixels with a preset value in the smallest rectangle serving as a candidate target is not less than a predetermined number; the fusion unit 1104B performs fusion on the obtained candidate targets to obtain a fusion target.

The counting unit 1106 includes:

The selecting unit 1106B finds a fusion target that satisfies a second preset condition from the obtained fusion targets as an object fusion target, wherein the second preset condition is determined by pixels with a preset value in the fusion target that is an object fusion target The ratio of the occupied area to the entire occupied area of the fusion target is greater than a predetermined ratio threshold; the matching unit 1106C, for each object fusion target in the sub-image block of the current image, obtained from the same sub-image block of at least the previous image In the fusion target, find the fusion target that has the largest overlapping area with the object fusion target and the maximum overlapping area exceeds the predetermined area threshold as the most overlapping fusion target of the object fusion target; the comparison unit 1106D, the sub-image block of the current image Each of the object fusion targets and at least the corresponding most overlapping fusion targets in the same sub-image block of the previous image are compared with the boundary line to decide whether to increment the statistical value, wherein, if the object in the sub-image block of the current image The statistic is incremented when a fused object crosses a boundary line without at least the corresponding most overlapping fused object in the same sub-image block of the previous image not crossing the same boundary line.

According to the image-based target object counting device of the present invention, a statistical area is defined in an image, and sub-image blocks corresponding to each boundary line of the statistical area are extracted, and a target object counting algorithm is executed for each sub-image block, and then each sub-image is The statistical results of blocks are combined to obtain the number of target objects entering the statistical area. Therefore, the degree of attention or interest of the corresponding product or location can be understood according to the statistical results. At the same time, the merged results are deduplicated to solve the problem of the same The target object is repeatedly counted, thereby improving the counting accuracy.

In addition, an embodiment of the present invention also provides an electronic device, which includes the above apparatus for counting target objects based on an image. In the specific implementation of the above-mentioned electronic device according to the embodiment of the present invention, the above-mentioned electronic device can be any one of the following devices: computer; tablet computer; personal digital assistant; multimedia playback device; mobile phone and electronic paper book, etc. Wait. Wherein, the electronic device has various functions and technical effects of the above-mentioned apparatus for counting objects based on an image, which will not be repeated here.

Each constituent unit, subunit, module, etc. of the image-based object counting apparatus according to the embodiment of the present invention may be configured by software, firmware, hardware or any combination thereof. In the case of realization by software or firmware, the program constituting the software or firmware can be installed from a storage medium or a network to a machine with a dedicated hardware structure, and when the machine is installed with various programs, it can execute the above-mentioned constituent units and sub-units. Various functions of the unit.

In addition, the present invention also proposes a program product storing machine-readable instruction codes. When the above instruction code is read and executed by a machine, the image-based target object counting method according to the embodiment of the present invention can be executed. Accordingly, various storage media such as magnetic disks, optical disks, magneto-optical disks, semiconductor memories, etc. for carrying such program products are also included in the disclosure of the present invention.

In addition, the methods in the various embodiments of the present invention are not limited to being executed in the time sequence described in the description or shown in the drawings, and may also be executed in other time sequences, in parallel or independently. Therefore, the execution order of the methods described in this specification does not limit the technical scope of the present invention.

In addition, obviously, each operation process of the above method according to the present invention can also be implemented in the form of computer executable programs stored in various machine-readable storage media.

Moreover, the purpose of the present invention can also be achieved in the following manner: the storage medium storing the above-mentioned executable program code is directly or indirectly provided to a system or device, and the computer or central processing unit in the system or device reads and Execute the above program code.

At this time, as long as the system or device has the function of executing the program, the embodiment of the present invention is not limited to the program, and the program can also be in any form, for example, an object program, a program executed by an interpreter, or a program provided to an operating system. script programs, etc.

The above-mentioned machine-readable storage media include, but are not limited to: various memories and storage units, semiconductor devices, magnetic disk units such as optical, magnetic and magneto-optical disks, and other media suitable for storing information, and the like.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (10)

1. An image-based target object counting method comprising: Extracting a plurality of sub-image blocks from the current image, each of the sub-image blocks is an area outside a boundary line of a predetermined statistical area and simultaneously bounded by the boundary line; For each sub-image block, obtain a foreground image of the sub-image block, and obtain a fusion target from the foreground image, each fusion target represents a passing target object; comparing fused objects obtained from a sub-image block in said current image and at least the same sub-image block in a previous image with boundary lines to perform object counting; A plurality of statistical results based on the sub-image blocks are fused to obtain the number of target objects entering the predetermined statistical area. 2. The image-based target object counting method according to claim 1, wherein when performing target object counting according to the sub-image blocks, each target object that crosses the boundary line and enters the predetermined counting area is recorded Information; Fusing the statistical results of multiple sub-image blocks includes: judging whether the same target object entering the predetermined statistical area at the same time is repeatedly counted according to the information of the target object; Filter the statistical results according to the judgment results. 3. The image-based target object counting method according to claim 2, wherein the information includes a statistical time point and a passing position, and the passing position is the angle position where two boundary lines of the predetermined statistical area intersect ; The judging whether the same target object entering the predetermined statistical area at the same time is repeatedly counted includes: If the recorded passing positions of the two target objects are the same, and the difference between the statistical time points of the two target objects is less than a preset time period, then it is determined that the two target objects enter the The same target object of the predetermined statistical area; The count of the target object whose statistical time point is later among the two target objects is ignored. 4. The image-based target object counting method according to claim 1, obtaining the fusion target comprises: Taking the connected area of pixels with a preset value in the foreground image and the smallest rectangle that satisfies a first preset condition as candidate targets, each candidate target represents a candidate for a passing target object, wherein the first preset It is assumed that the number of pixels having the preset value in the smallest rectangle as the candidate target is not less than a predetermined number; Fusion is performed on the obtained candidate objects to obtain the fused object. 5. The image-based target object counting method according to any one of claims 1 to 4, said performing target object number statistics, comprising: Find a fusion target that satisfies a preset condition from the obtained fusion targets as an object fusion target, wherein the preset condition is the area occupied by pixels with the preset value in the fusion target as the object fusion target The ratio to the entire occupied area of the fusion target is greater than a predetermined ratio threshold; For each object fusion target in the sub-image block of the current image, find the fusion target with the largest overlapping area with the object fusion target and the largest overlapping area from the fusion targets obtained in the same sub-image block of at least the previous image A fusion target exceeding a predetermined area threshold is taken as the most overlapping fusion target of the object fusion target; comparing each of the object fusion targets of a sub-image block of the current image and a corresponding most overlapping fusion target of at least the same sub-image block of a previous image with the boundary line to decide whether to increment the statistical value, Wherein, if the object fusion target in the sub-image block of the current image crosses the boundary line and the corresponding most overlapping fusion target in the same sub-image block of at least the previous image does not cross the same boundary line, the statistical value is incremented. 6. An image-based target object counting device comprising: The extraction unit is configured to extract a plurality of sub-image blocks from the current image, each sub-image block is an area outside a boundary line of a predetermined statistical area and at the same time bounded by the boundary line; A fused target acquisition unit, for each sub-image block, obtains a foreground image of the sub-image block, and obtains a fused target from the foreground image, and each fused target represents a passing target object; a counting unit for comparing the fusion target obtained from the sub-image block in the current image and at least the same sub-image block in the previous image with the boundary line to perform counting of the number of target objects; The statistical result fusion unit is configured to fuse a plurality of statistical results based on the sub-image blocks to obtain the number of target objects entering the predetermined statistical area. 7. The image-based target object counting device according to claim 6, said counting unit comprising: a recording unit, when performing counting of target objects according to said sub-image block, recording crosses said boundary line and enters said predetermined Information on each target object in the statistical area; The statistical result fusion unit includes: The judging unit, according to the information of the target object, judges whether the same target object entering the predetermined statistical area at the same time is counted repeatedly; The screening unit performs screening processing on the statistical results according to the judgment results. 8. The image-based target object counting device according to claim 7, wherein the information includes a statistical time point and a passing position, and the passing position is the angle position where two boundary lines of the predetermined statistical area intersect ; The judging unit is further configured to determine the two targets when it is judged that the recorded passing positions of the two target objects are the same, and the difference between the statistical time points of the two target objects is less than a preset time period. The object is the same target object entering the predetermined statistical area at the same time; The screening unit ignores counts of target objects whose statistical time point is later among the two target objects. 9. The image-based target object counting device according to claim 6, the fusion target acquisition unit comprising: The candidate target determination unit uses the connected area of pixels with preset values in the foreground image and the smallest rectangle that satisfies the first preset condition as candidate targets, and each candidate target represents a candidate for a passing target object, wherein, The first preset condition is that the number of pixels with the preset value in the smallest rectangle serving as the candidate target is not less than a predetermined number; A fusion unit is configured to perform fusion on the obtained candidate objects to obtain the fusion object. 10. The image-based target object counting device according to any one of claims 6 to 9, said counting unit comprising: The selection unit is used to find a fusion target that satisfies a second preset condition from the obtained fusion targets as an object fusion target, wherein the second preset condition is that the fusion target that is the object fusion target has the preset The ratio of the area occupied by the set pixel to the entire area occupied by the fusion target is greater than a predetermined ratio threshold; The matching unit, for each object fusion target in the sub-image block of the current image, finds out the fusion target with the largest overlapping area with the object fusion target from at least the fusion targets obtained in the same sub-image block of the previous image and the maximum The fusion target whose overlapping area exceeds the predetermined area threshold is used as the most overlapping fusion target of the object fusion target; a comparison unit for comparing each of the object fusion targets of the sub-image block of the current image and the corresponding most overlapping fusion targets in the same sub-image block of at least the previous image with the boundary line to determine whether to increment the statistical value, Wherein, if the object fusion target in the sub-image block of the current image crosses the boundary line and the corresponding most overlapping fusion target in the same sub-image block of at least the previous image does not cross the same boundary line, the statistical value is incremented.