An embedded real-time humanoid target automatic monitoring and recognition method

Abstract

The present invention discloses an embeddable real-time humanoid target automatic monitoring and identification method, which belongs to the technical field of humanoid target identification. The method comprises: establishing an embedding solution library based on the business needs of a platform party; obtaining monitoring demand information of a user, and extracting embedding demand features from the monitoring demand information; matching corresponding embedding solutions from the embedding solution library according to the embedding demand features; performing embedding adjustment of monitoring equipment according to the embedding solution and the monitoring demand information; setting a behavior monitoring list, and establishing a corresponding personnel identification model according to the behavior monitoring list; performing real-time identification of monitoring data through the personnel identification model to obtain corresponding personnel identification results; generating a behavior analysis map of the corresponding identified personnel according to the obtained personnel identification results; identifying the implementation behavior of the personnel according to the behavior analysis map; matching corresponding processing measures from a preset behavior measure library according to the implementation behavior, and processing according to the processing measures.

Description

Technical Field

[0001] The invention belongs to the technical field of humanoid target recognition, and in particular is an embedded real-time humanoid target automatic monitoring and recognition method. Background Art

[0002] With the rapid development of computer vision technology, humanoid target recognition is increasingly used in security, intelligent monitoring and other fields. However, existing humanoid target recognition systems often have problems such as low recognition accuracy and weak real-time performance. Especially in complex scenes, such as light changes and target occlusion, the recognition effect is often greatly reduced.

[0003] At present, although there are some real-time human target recognition systems on the market, most of them are based on large servers or high-end computers, and the equipment is expensive and inconvenient to carry. At the same time, these systems often need to transmit and process a large amount of data in the cloud, which not only increases the network burden, but also poses the risk of data leakage.

[0004] Based on this, the present invention provides an embedded real-time humanoid target automatic monitoring and recognition method. Summary of the invention

[0005] In order to solve the problems existing in the above-mentioned solutions, the present invention provides an embedded real-time humanoid target automatic monitoring and recognition method.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] An embedded real-time humanoid target automatic monitoring and recognition method, the method comprising:

[0008] Step 1: Establish an embedding solution library based on the platform's business needs, where the embedding solution is used to store various embedding solutions and the embedding classification set and device list corresponding to each embedding solution;

[0009] Furthermore, the method for establishing the embedding solution library includes:

[0010] Determine each device to be embedded based on business needs; set the corresponding embedding feature template according to the embedding requirements;

[0011] Acquire device information of each device to be embedded, extract features of each device information through the embedding feature template, and obtain embedding features of each device to be embedded;

[0012] Merge each embedded feature to obtain several embedded classification sets; integrate each device information to be embedded according to the obtained embedded classification sets to obtain a device list corresponding to each embedded classification set;

[0013] Set up corresponding embedding schemes according to the embedding classification set; integrate the embedding schemes, embedding classification sets and corresponding equipment lists to establish an embedding scheme library.

[0014] Furthermore, the method for merging the embedded features includes:

[0015] Identify each embedded feature item according to the embedded feature template; preset the interval classification range of each embedded feature item;

[0016] Set the corresponding embedding evaluation model according to the interval classification range of each embedding feature item;

[0017] Combining each embedded feature in pairs to obtain a number of comparison combinations, and integrating the two embedded feature item data corresponding to the embedded feature items in the comparison combination into feature item comparison data;

[0018] According to the formula Calculate the corresponding merged value; where: HD is the merged value; P(xi) is the embedded evaluation model;

[0019] Merge the two embedded features with a merge value of 0, and so on to obtain several merge sets; each merge set is marked as the corresponding embedded classification set.

[0020] Furthermore, the expression of the embedding evaluation model is ;

[0021] Wherein: i represents the corresponding embedded feature item, i=1, 2, ..., n, n is a positive integer; xi is the input data, which represents the feature item comparison data corresponding to the embedded feature item.

[0022] Step 2: Obtain the user's monitoring demand information, extract features from the monitoring demand information according to the preset embedding feature template, and obtain embedding demand features; input the obtained embedding demand features into the embedding solution library for feature matching to obtain the corresponding embedding solution; and perform embedding adjustment of the monitoring equipment according to the obtained embedding solution and the monitoring demand information;

[0023] Step 3: Setting a behavior monitoring list, which is used to count each target behavior; establishing a corresponding personnel identification model according to the behavior monitoring list;

[0024] Step 4: Use the personnel recognition model to identify the monitoring data in real time to obtain the corresponding personnel recognition results; generate a behavior analysis map of the corresponding identified personnel based on the obtained personnel recognition results;

[0025] Furthermore, the method for generating the behavior analysis graph includes:

[0026] According to each target behavior in the behavior monitoring list, a single curve graph template corresponding to each target behavior is established, wherein the horizontal axis of the single curve graph template is time, and the vertical axis is the behavior probability of the corresponding target behavior; and a correction function is set corresponding to each single curve graph template; the single curve graphs corresponding to each target behavior are integrated to obtain a behavior analysis graph template;

[0027] According to the personnel identification result, the behavior probability corresponding to each target behavior is identified, and the obtained behavior probability of each target behavior is input into the corresponding single curve graph template in the behavior analysis map template; the current behavior analysis map template is marked as a behavior analysis map.

[0028] Furthermore, when the result of person identification is that no person is identified, the behavior analysis graph is not generated.

[0029] Step 5: Conduct behavioral analysis based on the behavioral analysis graph to determine the corresponding implementation behavior of the identified person;

[0030] Furthermore, the method of performing behavior analysis according to the behavior analysis map includes:

[0031] Identify the single curve graph corresponding to each target behavior in the behavior analysis graph; fit each single curve graph template to obtain the behavior curve corresponding to each target behavior; mark the behavior curve as Gj(t), j represents the corresponding target behavior, j=1, 2, ..., m, m is a positive integer; identify the correction function corresponding to each single curve graph, and mark the correction function corresponding to each single curve graph as Wj(t);

[0032] Determine the reference time corresponding to each target behavior according to the behavior curve corresponding to each target behavior and the correction function; mark the obtained reference time as t0j;

[0033] According to the formula Calculate the single evaluation value of each target behavior; where: RYj is the single evaluation value of the corresponding target behavior;

[0034] According to the formula Calculate the behavior evaluation value of each target behavior; where: PLj is the behavior evaluation value of the corresponding target behavior;

[0035] The target behavior with the largest behavior evaluation value is selected as the implementation behavior of the identification personnel.

[0036] Furthermore, the method for determining the reference time includes:

[0037] According to the formula Determine the time evaluation value corresponding to each target behavior; where TPj is the time evaluation value of the corresponding target behavior, identify the times corresponding to the time evaluation value of 0, mark them as zero time, and select the zero time with the smallest time difference with the current time as the reference time.

[0038] Step 6: Establish a measures library based on the behavior monitoring list, and the measures library is used to store the processing measures corresponding to each target behavior; identify the corresponding implementation behavior in real time, match the corresponding processing measures from the behavior measures library according to the implementation behavior, and perform corresponding processing according to the obtained processing measures.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] By adopting embedded technology, it can be integrated into various devices, such as cameras, sensors, etc. This design makes the system highly flexible and portable, and can adapt to different application scenarios and needs. It can capture video streams in real time and monitor human targets in them. Through efficient image processing algorithms, the system can quickly and accurately identify targets, track and locate them, identify corresponding implementation behaviors, and perform corresponding processing based on the implementation behaviors to achieve intelligent management. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] 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.

[0042] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0043] 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.

[0044] like Figure 1 As shown, an embeddable real-time humanoid target automatic monitoring and recognition method includes:

[0045] Step 1: Obtain the business needs of the platform, which include data on fields, industries, directions, etc. in which the platform may apply the humanoid target automatic monitoring system and equipment during the business process; analyze the obtained business needs and establish an embedded solution library;

[0046] Methods for analyzing business needs and establishing an embedded solution library include:

[0047] Based on big data, we match and analyze business needs, obtain various devices to be embedded that meet business needs on the market, such as various cameras, sensors, etc.; that is, various related devices that need to be embedded that can meet the business requirements of the platform; set corresponding embedding feature templates according to embedding requirements, that is, set corresponding feature items according to various influencing factors that affect the embedding method, and integrate them into embedding feature templates for corresponding embedding feature extraction;

[0048] Obtain device information of each device to be embedded, extract features of each device information through an embedding feature template, and obtain embedding features of each device to be embedded;

[0049] Merge each embedded feature to obtain several embedded classification sets; integrate each device information to be embedded according to the obtained embedded classification sets to obtain a device list corresponding to each embedded classification set;

[0050] The platform sets the corresponding embedding scheme according to the embedding classification set; integrates each embedding scheme, embedding classification set and corresponding equipment list to establish an embedding scheme library.

[0051] Methods for merging embedded features include:

[0052] Identify each embedded feature item according to the embedded feature template; preset the interval classification range of each embedded feature item, that is, the embedded feature item data in the same interval classification range can be regarded as the same, and the same embedding method can be applied; the staff will set it according to the actual situation of the platform;

[0053] The corresponding embedding evaluation model is set according to the interval classification range of each embedded feature item. Specifically, it is established based on the existing recognition comparison algorithm, that is, the corresponding two embedded feature item data are compared to determine whether the same interval classification range is input. If so, the evaluation meets the embedding standard, otherwise, it is considered not to meet the embedding standard; the expression is ; Wherein: i represents the corresponding embedded feature item, i=1, 2, ..., n, n is a positive integer; xi is the input data, representing the embedded feature item data of the two comparisons of the corresponding embedded feature items;

[0054] Combine each embedded feature in pairs to obtain several comparison combinations, and integrate the two embedded feature item data corresponding to the corresponding embedded feature items in the comparison combination into feature item comparison data; i.e., the input data of the embedded evaluation model;

[0055] According to the formula Calculate the corresponding merged value; where: HD is the merged value; P(xi) is the embedded evaluation model;

[0056] Merge two embedded features with a merge value of 0, and so on to obtain several merge sets, that is, merge the embedded features with a merge value of 0 into one merge set; mark each merge set as the corresponding embedded classification set.

[0057] Step 2: Obtain the user's monitoring demand information, including monitoring demand, monitoring equipment details and other related data; extract the monitoring demand information according to the preset embedding feature template to obtain the corresponding embedding feature, which is marked as the embedding demand feature; input the obtained embedding demand feature into the embedding solution library for feature matching to obtain the corresponding embedding solution; adjust the embedding of the monitoring equipment according to the obtained embedding solution and monitoring demand information;

[0058] After obtaining the embedding plan, the staff will adjust the embedding plan according to the actual needs of the user, and after the negotiation is completed, the embedding adjustment of the monitoring equipment will be carried out.

[0059] Step 3: Set up a behavior monitoring list according to user needs. The behavior monitoring list records the behaviors of each person that the user needs to monitor and identify, and marks them as target behaviors; such as normal falls, walking, running, sudden illness falls, entering prohibited areas, and other behaviors; based on the obtained behavior monitoring list, the platform establishes a corresponding personnel recognition model based on a neural network such as a CNN network or a DNN network. The input data is the monitoring data, that is, the surveillance video; the output data is the personnel recognition result, which includes the probability of each target behavior corresponding to the person not being recognized, the person being recognized, and the recognized person; because for behavior recognition, the same action may correspond to various target behaviors, such as a common fall, which may be a normal fall and the person can stand up soon afterwards, or it may be a fall after a sudden illness and the person cannot get up for a long time after falling; the specific needs need to be evaluated in combination with the corresponding actual performance characteristics; the corresponding training set is established manually for training, and the training set includes input data and output data;

[0060] Step 4: Use the personnel recognition model to identify the monitoring data in real time to obtain the corresponding personnel recognition results; generate a behavior analysis map of the corresponding identified personnel based on the obtained personnel recognition results;

[0061] The behavior analysis graph is established according to each target behavior in the behavior monitoring list. Each target behavior corresponds to a single curve graph, the horizontal axis of the single curve graph is time, and the vertical axis is the behavior probability corresponding to the target behavior; and a time correction function is set for the single curve graph corresponding to each target behavior, which is mainly used for probability correction for the floating change of its behavior probability with the cumulative change over time, because the behavior probability identified by the personnel identification model is an instantaneous probability, and for normal falls, as the fall time increases, its behavior probability gradually decreases, while the sudden illness and fall will increase; specifically, the platform obtains a large amount of historical data of each target behavior, and statistically analyzes the conversion of its true probability and behavior probability over time, and after simulation, determines the correction function corresponding to each target behavior;

[0062] If the result of person identification is that the person is not identified, there is no need to generate a behavior analysis map; because the behavior analysis map is for the corresponding identified person and serves as the behavior tracking data of the identified person.

[0063] Step 5: Conduct behavioral analysis based on the behavioral analysis graph to determine the corresponding implementation behavior of the identified person;

[0064] Methods for behavioral analysis based on behavioral analysis maps include:

[0065] Identify the single curve graph corresponding to each target behavior in the behavior analysis graph; fit each single curve graph template to obtain the behavior curve corresponding to each target behavior; mark the behavior curve as Gj(t), j represents the corresponding target behavior, j=1, 2, ..., m, m is a positive integer; identify the correction function corresponding to each single curve graph, and mark the correction function corresponding to each single curve graph as Wj(t);

[0066] According to the formula Determine the time evaluation value corresponding to each target behavior; where TPj is the time evaluation value corresponding to the target behavior, identify each time corresponding to the time evaluation value of 0, mark it as zero time, select the zero time with the smallest time difference with the current time as the reference time; mark the obtained reference time as t0j;

[0067] According to the formula Calculate the single evaluation value of each target behavior; where: RYj is the single evaluation value of the corresponding target behavior;

[0068] According to the formula Calculate the behavior evaluation value of each target behavior; where: PLj is the behavior evaluation value of the corresponding target behavior;

[0069] The target behavior with the largest behavior evaluation value is selected as the implementation behavior of the identification personnel.

[0070] Step 6: Set the corresponding processing measures for each target behavior according to the behavior monitoring list and user needs, such as no operation, various corresponding warning methods, etc.; organize the set processing measures and establish a corresponding behavior measure library; identify the corresponding implementation behavior of the identification personnel in real time, match the corresponding processing measures from the behavior measure library according to the implementation behavior, and perform corresponding processing according to the obtained processing measures.

[0071] 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.

[0072] 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. An embedded real-time humanoid target automatic monitoring and recognition method, characterized in that: Methods include: Step 1: Establish an embedding solution library based on the platform's business needs, where the embedding solution is used to store various embedding solutions and the embedding classification set and device list corresponding to each embedding solution; Step 2: Obtain the user's monitoring demand information, extract features from the monitoring demand information according to a preset embedded feature template, and obtain embedded demand features; Input the obtained embedding requirement features into the embedding solution library for feature matching to obtain the corresponding embedding solution; perform embedding adjustment of the monitoring device according to the obtained embedding solution and monitoring requirement information; Step 3: Setting a behavior monitoring list, which is used to count each target behavior; establishing a corresponding personnel identification model according to the behavior monitoring list; Step 4: Use the personnel recognition model to identify the monitoring data in real time to obtain the corresponding personnel recognition results; generate a behavior analysis map of the corresponding identified personnel based on the obtained personnel recognition results; Step 5: Conduct behavioral analysis based on the behavioral analysis graph to determine the corresponding implementation behavior of the identified person; Step 6: Establish a measure library based on the behavior monitoring list, and the measure library is used to store the treatment measures corresponding to each target behavior; Identify the corresponding implementation behavior in real time, match the corresponding processing measures from the behavior measure library according to the implementation behavior, and perform corresponding processing according to the obtained processing measures.

2. The method for automatically monitoring and identifying a humanoid target in real time according to claim 1 is characterized in that: The method for establishing the embedding solution library includes: Determine each device to be embedded based on business needs; set the corresponding embedding feature template according to the embedding requirements; Acquire device information of each device to be embedded, extract features of each device information through the embedding feature template, and obtain embedding features of each device to be embedded; Merge each embedded feature to obtain several embedded classification sets; integrate each device information to be embedded according to the obtained embedded classification sets to obtain a device list corresponding to each embedded classification set; Set up corresponding embedding schemes according to the embedding classification set; integrate the embedding schemes, embedding classification sets and corresponding equipment lists to establish an embedding scheme library.

3. The method for automatically monitoring and identifying a humanoid target in real time according to claim 2 is characterized in that: Methods for merging embedded features include: Identify each embedded feature item according to the embedded feature template; preset the interval classification range of each embedded feature item; Set the corresponding embedding evaluation model according to the interval classification range of each embedding feature item; Combining each embedded feature in pairs to obtain a number of comparison combinations, and integrating the two embedded feature item data corresponding to the embedded feature items in the comparison combination into feature item comparison data; According to the formula Calculate the corresponding merged value; where: HD is the merged value; P(xi) is the embedded evaluation model; Merge the two embedded features with a merge value of 0, and so on, to obtain several merge sets; mark each merge set as the corresponding embedded classification set; The expression of the embedding evaluation model is ; Wherein: i represents the corresponding embedded feature item, i=1, 2, ..., n, n is a positive integer; xi is the input data, which represents the feature item comparison data corresponding to the embedded feature item.

4. The method for automatically monitoring and identifying a humanoid target in real time according to claim 1, characterized in that: The method for generating the behavior analysis map includes: According to each target behavior in the behavior monitoring list, a single curve graph template corresponding to each target behavior is established, wherein the horizontal axis of the single curve graph template is time, and the vertical axis is the behavior probability of the corresponding target behavior; and a correction function is set corresponding to each single curve graph template; the single curve graphs corresponding to each target behavior are integrated to obtain a behavior analysis graph template; According to the personnel identification result, the behavior probability corresponding to each target behavior is identified, and the obtained behavior probability of each target behavior is input into the corresponding single curve graph template in the behavior analysis map template; the current behavior analysis map template is marked as a behavior analysis map.

5. The method for automatically monitoring and identifying a humanoid target in real time according to claim 4 is characterized in that: When the result of person recognition is that no person is recognized, the behavior analysis graph is not generated.

6. The method for automatically monitoring and identifying a humanoid target in real time according to claim 1, characterized in that: Methods for behavioral analysis based on behavioral analysis maps include: Identify the single curve graph corresponding to each target behavior in the behavior analysis graph; fit each single curve graph template to obtain the behavior curve corresponding to each target behavior; mark the behavior curve as Gj(t), j represents the corresponding target behavior, j=1, 2, ..., m, m is a positive integer; identify the correction function corresponding to each single curve graph, and mark the correction function corresponding to each single curve graph as Wj(t); Determine the reference time corresponding to each target behavior according to the behavior curve corresponding to each target behavior and the correction function; mark the obtained reference time as t0j; According to the formula Calculate the single evaluation value of each target behavior; where: RYj is the single evaluation value of the corresponding target behavior; According to the formula Calculate the behavior evaluation value of each target behavior; where: PLj is the behavior evaluation value of the corresponding target behavior; The target behavior with the largest behavior evaluation value is selected as the implementation behavior of the identification personnel.

7. The method for automatically monitoring and identifying a humanoid target in real time according to claim 6 is characterized in that: Methods for determining the benchmark time include: According to the formula Determine the time evaluation value corresponding to each target behavior; where TPj is the time evaluation value of the corresponding target behavior, identify the times corresponding to the time evaluation value of 0, mark them as zero time, and select the zero time with the smallest time difference with the current time as the reference time.

Images