A video target detection and tracking method and system based on artificial intelligence

Abstract

The invention discloses a video target detection and tracking method and system based on artificial intelligence, which belongs to the field of video monitoring technology. The method comprises: step one: marking a video target, extracting a target image of the video target, and performing corresponding verification and correction; step two: grayscale processing the target image to obtain a target grayscale image, analyzing the target grayscale image, and obtaining an initial recognition feature of the video target; step three: identifying each pixel block in the initial recognition feature, determining a corresponding positioning block according to a block attribute corresponding to each pixel block, and the positioning block is used to determine the pixel block of the video target or a combination block between each pixel block; step four: acquiring a monitoring video in real time, grayscale processing the monitoring video, obtaining a corresponding monitoring grayscale image, and determining a corresponding positioning matching block in the monitoring grayscale image according to the positioning block; step five: identifying the corresponding video target according to the positioning matching block and the initial recognition feature.

Description

Technical Field

[0001] The present invention belongs to the technical field of video surveillance, and in particular relates to a video target detection and tracking method and system based on artificial intelligence. Background Art

[0002] With the rapid development of information technology, video surveillance technology has been widely used in various fields. Traditional video surveillance mainly relies on manual monitoring, which is not only inefficient but also easy to miss key information. In recent years, with the continuous advancement of artificial intelligence technology, especially the development of deep learning and computer vision technology, automatic video target detection and tracking have become possible. Therefore, how to perform video target detection and tracking better and faster is the direction that needs to be solved and improved at present; based on this, the present invention provides a video target detection and tracking method and system based on artificial intelligence. Summary of the invention

[0003] In order to solve the problems existing in the above scheme, the present invention provides a video target detection and tracking method and system based on artificial intelligence.

[0004] The purpose of the present invention can be achieved through the following technical solutions:

[0005] A video target detection and tracking method based on artificial intelligence, the method comprising:

[0006] Step 1: Mark the video target, extract the target image of the video target, and perform corresponding verification and correction;

[0007] Step 2: grayscale processing is performed on the target image to obtain a target grayscale image, and the target grayscale image is analyzed to obtain initial recognition features of the video target;

[0008] Furthermore, the method for analyzing the target grayscale image includes:

[0009] Identify the grayscale value of each pixel in the target grayscale image, merge the grayscale values ​​of adjacent pixels, and obtain a number of pixel blocks and block attributes corresponding to each pixel block; mark each pixel block in the target grayscale image according to the block attributes, and mark the corresponding block attributes; segment the target grayscale image to obtain initial recognition features corresponding to the video target.

[0010] Furthermore, the method of merging according to the grayscale values ​​between adjacent pixels includes:

[0011] Step SA1: Mark each pixel as a single sample, identify the grayscale difference between the grayscale values ​​corresponding to each adjacent single sample, merge the adjacent single samples whose grayscale difference is less than the threshold X1 to obtain a merged sample, and calculate the grayscale value corresponding to the merged sample according to the merging formula;

[0012] Step SA2: determine the sample to be merged of the merged sample; calculate the grayscale difference between the merged sample and the sample to be merged, merge the merged sample whose grayscale difference is less than the threshold X1 with the sample to be merged to obtain a new merged sample, and calculate the grayscale value of the new merged sample according to the merging formula;

[0013] Step SA3: looping step SA2 until there is no sample to be merged that meets the merging requirements, marking the corresponding merged sample as a pixel block;

[0014] Step SA4: Determine whether there is a merged block. If there is a merged block, return to step SA3; if there is no merged block, identify the corresponding part of each pixel block, determine the corresponding stable value according to the identified part, and identify the block shape, position relationship and grayscale value of each pixel block; integrate the obtained stable value, part, block shape, position relationship and grayscale value into the block attribute of the corresponding pixel block.

[0015] Furthermore, the combined formula is: Where: Hd is the grayscale value of the merged sample; HB is the sum of the grayscale values ​​corresponding to each single sample in the merged sample; L is the number of single samples in the merged sample.

[0016] Furthermore, the method for determining the stable value includes:

[0017] Matching the initial values ​​corresponding to each part according to the identified part;

[0018] Obtain current environmental information, analyze each part according to the obtained environmental information, and obtain the adjustment coefficient corresponding to each part;

[0019] The corresponding stable value is calculated according to the formula WD=CZ×τ; where: WD is the stable value; CZ is the initial value; τ is the adjustment coefficient.

[0020] Step 3: Identify each pixel block in the initial recognition feature, and determine the corresponding positioning block according to the block attribute corresponding to each pixel block, and the positioning block is used to determine the pixel block of the video target or the combination block between the pixel blocks;

[0021] The method for determining the positioning block includes:

[0022] Identifying block attributes of each pixel block, the block attributes including stability value, location, block shape, position relationship and grayscale value;

[0023] Determine each first combination according to the positional relationship between each pixel block; identify each pixel block in the first combination and mark it as a unit block;

[0024] identifying the combined shape data of the first combination; calculating the first shape value;

[0025] Identify the stable values corresponding to each unit block, and select the lowest stable value in each unit block as the representative stable value of the first combination;

[0026] Calculate the corresponding positioning evaluation value according to the formula QYU = (b1 × WDB) × (b2 × BA);

[0027] In the formula: QYU is the positioning evaluation value; b1 and b2 are both proportionality coefficients, and the value range is 0 < b1 ≤ 1, 0 < b2 ≤ 1; WDB is the representative stable value; BA is the first shape value;

[0028] Select the first combination with the largest positioning evaluation value as the positioning block.

[0029] Furthermore, the calculation method of the first shape value includes:

[0030] Obtain the preset environmental background characteristics; determine the reference background according to the environmental background characteristics and the combined shape data, identify the reference similarity value between the reference background and the combined shape data, and count the reference probability of the reference background;

[0031] Conduct boundary assimilation evaluation on the combined shape data and the environmental background characteristics according to the preset assimilation evaluation model, and the expression of the assimilation evaluation model is In the formula: x is the input data;

[0032] According to the formula Calculate the corresponding first shape value; in the formula: BA is the first shape value; SL is the reference similarity value; gL is the reference probability.

[0033] Step 4: Obtain the monitoring video in real time, perform grayscale processing on the monitoring video to obtain the corresponding monitoring grayscale image, and determine the corresponding positioning matching block according to the positioning block in the monitoring grayscale image;

[0034] Furthermore, the method for determining the positioning matching block includes:

[0035] Identify the grayscale values of the pixels in the monitoring grayscale image, merge them according to the grayscale values between adjacent pixels to obtain several monitoring blocks and the corresponding grayscale values and block shapes of each monitoring block;

[0036] Identify the grayscale values and block shapes of the pixel blocks in the positioning block, screen the monitoring blocks according to the block shapes of the pixel blocks to obtain the screened monitoring grayscale image;

[0037] Traverse and move the positioning block on the monitoring grayscale image, calculate the positioning matching value in real time, compare the obtained positioning matching values, and determine the positioning matching block.

[0038] Furthermore, the calculation method of the positioning matching value includes:

[0039] The gray value of the pixel block in the positioning block is marked as hi, where i represents the corresponding pixel block, i=1, 2, ..., n, and n is a positive integer;

[0040] Identify the monitoring blocks corresponding to each pixel block on the monitoring grayscale image, and mark the grayscale value of the corresponding monitoring block as ki;

[0041] According to the formula Calculate the corresponding positioning matching value; where: DPW is the positioning matching value.

[0042] Step 5: Identify the corresponding video target based on the positioning matching block and the initial recognition features.

[0043] A video target detection and tracking system based on artificial intelligence, including a target analysis module, a positioning module and a tracking module;

[0044] The target analysis module is used to analyze the marked video target and determine the corresponding initial recognition features;

[0045] The positioning module is used to determine the positioning block corresponding to the video target according to the initial recognition feature;

[0046] The tracking module is used to identify and track video targets, obtain surveillance videos in real time, perform grayscale processing on the surveillance videos, obtain corresponding surveillance grayscale images, determine corresponding positioning matching blocks in the surveillance grayscale images according to the positioning blocks; and identify corresponding video targets according to the positioning matching blocks and initial recognition features.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The present invention realizes real-time recognition and tracking of video targets, especially by quickly identifying the corresponding positioning matching blocks through the positioning blocks, thereby realizing fast and efficient positioning of video targets; and during the operation process, very few models and algorithms with high resource occupancy are configured, which facilitates fast operation. It realizes automatic real-time detection and tracking of targets in video streams without manual intervention, thereby greatly improving monitoring efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 This is a principle block diagram of the present invention. DETAILED DESCRIPTION

[0051] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] like Figure 1 As shown, a video target detection and tracking method based on artificial intelligence includes:

[0053] Step 1: Mark the video target that needs to be detected and tracked, extract the target image of the video target, and perform corresponding verification and correction, that is, show the extracted target image to the management personnel to determine whether it is the complete image marked for it. If not, make corresponding adjustments, such as cropping, etc.;

[0054] Step 2: grayscale processing is performed on the verified and adjusted target image to obtain a target grayscale image, and the target grayscale image is analyzed to obtain the initial recognition features of the video target;

[0055] Methods for analyzing the target grayscale image include:

[0056] Identify the grayscale value of each pixel in the target grayscale image, merge the grayscale values ​​of adjacent pixels, and obtain several pixel blocks and the block attributes corresponding to each pixel block; mark each pixel block in the target grayscale image according to the block attributes, and mark the corresponding block attributes; segment the target grayscale image, that is, segment the marked target grayscale image corresponding to the video target; obtain the initial recognition features corresponding to the video target.

[0057] Methods for merging pixels based on the grayscale values ​​of adjacent pixels include:

[0058] Step SA1: Mark each pixel as a single sample, identify the grayscale difference between the grayscale values ​​corresponding to each adjacent single sample, merge the adjacent single samples whose grayscale difference is less than the threshold X1 to obtain a merged sample, and calculate the grayscale value corresponding to the merged sample according to the merging formula;

[0059] The combined formula is: Where: Hd is the grayscale value of the merged sample; HB is the sum of the grayscale values ​​corresponding to each single sample in the merged sample; L is the number of single samples in the merged sample.

[0060] Step SA2: Determine the sample to be merged of the merged sample, that is, each single sample adjacent to the merged sample and the merged sample; calculate the grayscale difference between the merged sample and the sample to be merged, merge the merged sample whose grayscale difference is less than the threshold X1 with the sample to be merged to obtain a new merged sample, and calculate the grayscale value of the new merged sample according to the merging formula;

[0061] Step SA3: loop step SA2 until there is no sample to be merged that meets the merging requirement, and mark the corresponding merged sample as a pixel block; the merging requirement is that the grayscale difference is less than the threshold value X1; this step will stop when one pixel block is determined and enter the next step; therefore, there may be other merged blocks;

[0062] Step SA4: Determine whether there is a merged block. If there is a merged block, return to step SA3; if there is no merged block, it means that all merges have been completed. If there is no merged block, identify the corresponding part of each pixel block, determine the corresponding stable value according to the identified part, and identify the block shape, position relationship and grayscale value of each pixel block. The position relationship refers to the connection and adjacent relationship; integrate the obtained stable value, part, block shape, position relationship and grayscale value into the block attribute of the corresponding pixel block.

[0063] The methods for determining the stable value include:

[0064] Identify the parts corresponding to each pixel block, such as coat, palm, hair, face, eyes and other relevant parts for a person; the specific parts are determined according to the actual situation;

[0065] Count the various parts that may be encountered in the video detection and tracking application, count the change probability of each part, such as the probability of changing clothes, taking off clothes, etc., count the corresponding probability values, that is, remove the percentage sign; mark it as the initial value; and then match it according to the identified parts;

[0066] Obtain current environmental information, such as rainy days, snowy days, high temperatures, etc.; analyze each part according to the obtained environmental information to obtain the corresponding adjustment coefficient of each part, that is, set the corresponding adjustment coefficient according to the influence of the current environment; specifically, establish a corresponding intelligent model based on CNN network or DNN network, and establish a corresponding training set for training by artificial means. The training set includes input data and output data. The input data is the part and environmental information; the output data is the adjustment coefficient; analyze the intelligent model after successful training to obtain the corresponding adjustment coefficient;

[0067] The corresponding stable value is calculated according to the formula WD=CZ×τ; where: WD is the stable value; CZ is the initial value; τ is the adjustment coefficient.

[0068] Step 3: Identify each pixel block in the initial recognition feature, and determine the corresponding positioning block according to the block attribute corresponding to each pixel block. The positioning block is used to quickly determine the pixel block of the video target or the combination block between the pixel blocks;

[0069] The method for determining the positioning block includes:

[0070] Identify the block attributes of each pixel block, the block attributes include stability value, location, block shape, position relationship and gray value;

[0071] Determine various pixel block combinations according to the positional relationship between the pixel blocks, mark them as a first combination, and a single one may also be the first combination; identify the pixel blocks in the first combination, mark them as unit blocks;

[0072] Identify the combined shape data of the first combination, including the overall boundary shape, area, shape, area, grayscale value and other data of each unit block of the combination; obtain preset environmental background features, which are used to represent various shape data that may exist in the current environment. Because the environmental background under the monitoring background is basically fixed, controllable and predictable, the corresponding environmental background features can be preset according to the actual situation; establish a corresponding regional background feature map in combination with the corresponding regional map for subsequent direct matching according to the video target;

[0073] Determine the background most similar to the combined shape data according to the preset environmental background features, mark it as a reference background, identify the similarity between the reference background and the combined shape data, mark it as a benchmark similarity value, and calculate the probability of occurrence of the reference background according to historical data over a period of time, mark it as a benchmark probability;

[0074] The boundary assimilation evaluation is performed based on the combined shape data and the environmental background characteristics, that is, whether the unit blocks located at the boundary in the first combination are likely to assimilate with the environment. For example, if they are all positive red, the grayscale values ​​are not much different, which is easy to be compatible with the environment, but the specificity is insufficient, which is not conducive to rapid identification from the environment; a grayscale value difference interval is preset, and the assimilation evaluation standard is set according to the grayscale value difference interval and the occurrence probability of the background that meets the grayscale value difference interval, that is, the grayscale value difference interval must be reached and the corresponding occurrence probability must be higher than to be considered to meet the assimilation evaluation standard; according to the assimilation evaluation standard, the corresponding assimilation evaluation model is established, and the expression is: Where: x is the input data, and is the combined shape data and environmental background features;

[0075] According to the formula Calculate the corresponding first shape value; where: BA is the first shape value; SL is the benchmark similarity value; gL is the benchmark probability.

[0076] Identify the stable values corresponding to each unit block, and select the lowest stable value in each unit block as the representative stable value of the first combination; mark the representative stable value as WDB;

[0077] Calculate the corresponding positioning evaluation value according to the formula QYU=(b1×WDB)×(b2×BA); where: QYU is the positioning evaluation value; b1 and b2 are both proportionality coefficients, and the value range is 0 < b1 ≤ 1, 0 < b2 ≤ 1; WDB is the representative stable value; BA is the first shape value.

[0078] Select the first combination with the largest positioning evaluation value as the positioning block.

[0079] The positioning block is preferably a pixel block; it is convenient for identification and positioning; it is best not to exceed 5 pixel blocks. The fewer the number of pixel blocks, the higher the recognition efficiency and recognition accuracy.

[0080] Step 4: Obtain the monitoring video in real time, perform gray processing on the current monitoring screen to obtain a monitoring gray image, identify the gray values of each pixel in the monitoring gray image, and merge them according to the gray values between adjacent pixels to obtain several pixel blocks and the corresponding gray values and block shapes of each pixel block; mark the pixel blocks in the monitoring gray image as monitoring blocks;

[0081] Identify the gray values and block shapes of the pixel blocks corresponding to the positioning block, and screen the monitoring blocks according to the block shapes of the pixel blocks, that is, eliminate the monitoring blocks corresponding to the shapes that the pixel blocks will not become according to the actual situation, and establish a corresponding recognition evaluation model according to basic common sense and existing technologies for evaluation;

[0082] Mark the gray value of the pixel block in the positioning block as hi, where i represents the corresponding pixel block, i = 1, 2, ……, n, and n is a positive integer;

[0083] Traverse and move the positioning block on the monitoring gray image, calculate the positioning matching value in real time. If there are corresponding monitoring blocks to be eliminated, skip directly and do not calculate the positioning matching value, which is equivalent to having a hole on the monitoring gray image; compare the obtained positioning matching values to determine the positioning matching block, that is, the combination of monitoring blocks corresponding to the positioning block with the smallest positioning matching value; the positioning matching block is the image block on the monitoring gray image and is also part of the video target.

[0084] The calculation method of the positioning matching value includes:

[0085] Identify the monitoring blocks corresponding to each pixel block on the monitoring grayscale image, and mark the grayscale value of the corresponding monitoring block as ki, indicating that it corresponds to the corresponding pixel block; in practical applications, the corresponding monitoring blocks are generally quickly matched with the help of the established corresponding intelligent model, that is, the corresponding monitoring block is directly determined during the movement of the positioning block, and then direct calculation is performed based on the corresponding grayscale value to improve efficiency; specifically, the corresponding intelligent model can be established according to the existing technology, such as establishing a corresponding intelligent model based on a neural network.

[0086] According to the formula Calculate the corresponding positioning matching value; where: DPW is the positioning matching value.

[0087] Step 5: Identify the corresponding video target based on the positioning matching block and the initial recognition features.

[0088] The present invention realizes real-time recognition and tracking of video targets, especially by quickly identifying the corresponding positioning matching blocks through the positioning blocks, thereby realizing fast and efficient positioning of video targets; and during the operation process, very few models and algorithms with high resource occupancy are configured, which facilitates fast operation. It realizes automatic real-time detection and tracking of targets in video streams without manual intervention, thereby greatly improving monitoring efficiency.

[0089] A video target detection and tracking system based on artificial intelligence, including a target analysis module, a positioning module and a tracking module;

[0090] The target analysis module is used to analyze the marked video target and determine the corresponding initial recognition features.

[0091] The positioning module is used to determine the positioning block corresponding to the video target according to the initial recognition feature.

[0092] The tracking module is used to identify and track video targets, obtain surveillance videos in real time, perform grayscale processing on the surveillance videos, obtain corresponding surveillance grayscale images, determine corresponding positioning matching blocks in the surveillance grayscale images according to the positioning blocks; and identify corresponding video targets according to the positioning matching blocks and initial recognition features.

[0093] The above formulas are all calculated by removing dimensions and taking numerical values. The formula is a formula that is closest to the actual situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained by simulating a large amount of data.

[0094] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A video target detection and tracking method based on artificial intelligence, characterized in that: The method includes: Step 1: Mark the video target, extract the target image of the video target, and perform corresponding verification and correction; Step 2: Perform grayscale processing on the target image to obtain a target grayscale image, and analyze the target grayscale image to obtain initial recognition features of the video target; Step 3: Identify each pixel block in the initial recognition features, determine a corresponding positioning block according to the block attribute corresponding to each pixel block, and the positioning block is used to determine the pixel block of the video target or the combined block between each pixel block; Step 4: Obtain the monitoring video in real time, perform grayscale processing on the monitoring video to obtain a corresponding monitoring grayscale image, and determine a corresponding positioning matching block in the monitoring grayscale image according to the positioning block; Step 5: Identify the corresponding video target according to the positioning matching block and the initial recognition features; The method for determining the positioning block includes: Identify the block attributes of each pixel block, and the block attributes include a stability value, a part, a block shape, a positional relationship, and a grayscale value; Determine each first combination according to the positional relationship between each pixel block; identify each pixel block in the first combination and mark it as a unit block; Identify the combined shape data of the first combination; calculate the first shape value; Identify the stability value corresponding to each unit block, and select the lowest stability value among each unit block as the representative stability value of the first combination; Calculate the corresponding positioning evaluation value according to the formula QYU=(b1×WDB)×(b2×BA); In the formula: QYU is the positioning evaluation value; b1 and b2 are both proportionality coefficients, and the value range is 0 < b1 ≤ 1, 0 < b2 ≤ 1; WDB is the representative stability value; BA is the first shape value; Select the first combination with the largest positioning evaluation value as the positioning block; The method for calculating the first shape value includes: Obtain the preset environmental background feature; determine the reference background according to the environmental background feature and the combined shape data, identify the reference similarity value between the reference background and the combined shape data, and count the reference probability of the reference background; The boundary assimilation evaluation of the combined shape data and environmental background features is performed according to the preset assimilation evaluation model. The expression of the assimilation evaluation model is: ;Where: x is the input data; According to the formula Calculate the corresponding first shape value; where BA is the first shape value; SL is the benchmark similarity value; gL is the benchmark probability; The method for determining the positioning matching block includes: Identify the grayscale value of each pixel in the monitoring grayscale image, merge according to the grayscale value between each adjacent pixel to obtain several monitoring blocks and the grayscale value and block shape corresponding to each monitoring block; Identify the grayscale value and block shape of each pixel block in the positioning block, and screen each monitoring block according to the block shape of the pixel block to obtain a screened monitoring grayscale image; Traverse and move the positioning block on the monitoring grayscale image, calculate the positioning matching value in real time, compare the obtained positioning matching values, and determine the positioning matching block; The method for calculating the positioning matching value includes: Mark the grayscale value of the pixel block in the positioning block as hi, where i represents the corresponding pixel block, i = 1, 2,..., n, and n is a positive integer; Identify the monitoring block corresponding to each pixel block on the monitoring grayscale image, and mark the grayscale value of the corresponding monitoring block as ki; According to the formula Calculate the corresponding positioning matching value; where: DPW is the positioning matching value.

2. The video target detection and tracking method based on artificial intelligence according to claim 1 is characterized in that: The method for analyzing the target grayscale image includes: Identify the grayscale value of each pixel in the target grayscale image, merge the grayscale values ​​of adjacent pixels, and obtain a number of pixel blocks and block attributes corresponding to each pixel block; mark each pixel block in the target grayscale image according to the block attributes, and mark the corresponding block attributes; segment the target grayscale image to obtain initial recognition features corresponding to the video target.

3. The video target detection and tracking method based on artificial intelligence according to claim 2 is characterized in that: Methods for merging pixels based on the grayscale values ​​of adjacent pixels include: Step SA1: Mark each pixel as a single sample, identify the grayscale difference between the grayscale values ​​corresponding to each adjacent single sample, merge the adjacent single samples whose grayscale difference is less than the threshold X1 to obtain a merged sample, and calculate the grayscale value corresponding to the merged sample according to the merging formula; Step SA2: determine the sample to be merged of the merged sample; calculate the grayscale difference between the merged sample and the sample to be merged, merge the merged sample whose grayscale difference is less than the threshold X1 with the sample to be merged to obtain a new merged sample, and calculate the grayscale value of the new merged sample according to the merging formula; Step SA3: looping step SA2 until there is no sample to be merged that meets the merging requirements, marking the corresponding merged sample as a pixel block; Step SA4: Determine whether there is a merged block. If there is a merged block, return to step SA3; if there is no merged block, identify the corresponding part of each pixel block, determine the corresponding stable value according to the identified part, and identify the block shape, position relationship and grayscale value of each pixel block; integrate the obtained stable value, part, block shape, position relationship and grayscale value into the block attribute of the corresponding pixel block.

4. The video target detection and tracking method based on artificial intelligence according to claim 3 is characterized in that: The combined formula is: ; Where: Hd is the grayscale value of the merged sample; HB is the sum of the grayscale values ​​corresponding to each single sample in the merged sample; L is the number of single samples in the merged sample.

5. The video target detection and tracking method based on artificial intelligence according to claim 3 is characterized in that: The methods for determining the stable value include: Matching the initial values ​​corresponding to each part according to the identified part; Obtain current environmental information, analyze each part according to the obtained environmental information, and obtain the adjustment coefficient corresponding to each part; Calculate the corresponding stable value according to the formula WD=CZ×τ; Where: WD is the stable value; CZ is the initial value; τ is the adjustment coefficient.

6. A video target detection and tracking system based on artificial intelligence, characterized in that: Execute the video target detection and tracking method based on artificial intelligence as described in any one of claims 1 to 5, including a target analysis module, a positioning module and a tracking module; The target analysis module is used to analyze the marked video target and determine the corresponding initial recognition features; The positioning module is used to determine the positioning block corresponding to the video target according to the initial recognition feature; The tracking module is used to identify and track video targets, obtain surveillance videos in real time, perform grayscale processing on the surveillance videos, obtain corresponding surveillance grayscale images, determine corresponding positioning matching blocks in the surveillance grayscale images according to the positioning blocks; and identify corresponding video targets according to the positioning matching blocks and initial recognition features.

Images