Anti-theft method and system based on AI intelligence

Abstract

The invention discloses an anti-theft method and system based on AI intelligence. The method comprises the following steps: acquiring a monitoring video shot by a camera in real time; video stream decoding is carried out on the monitoring video, AI intelligent analysis is carried out on a monitoring object in the monitoring video, and the AI intelligent analysis comprises cross-camera tracking, human body attribute recognition and / or suspicious action recognition; and if the AI intelligent analysis finds that the monitoring object is stolen and damaged, an alarm is given to the monitoring object on the spot and a worker in the monitoring room is reminded to pay attention. According to the invention, monitoring video pictures are analyzed intelligently through AI, theft and loss behaviors are found in time, workers are assisted to carry out safety supervision, the working efficiency is improved, and the labor cost of monitoring work is reduced.

Description

Technical Field

[0001] The present invention belongs to the field of AI intelligent anti-theft technology, and specifically relates to an anti-theft method and system based on AI intelligence. Background Art

[0002] With the development of information technology, the reduction of hardware prices and network bandwidth costs, and the upgrade of network bandwidth, video surveillance systems have become a necessary security measure in various scenarios. Chain convenience stores have many stores, and each store is equipped with multiple surveillance cameras. There are too many surveillance videos, and the existing staff cannot maintain full attention to observe the video content throughout the day, and cannot observe all abnormal behaviors in time. In addition, the video images that a single staff member can observe are limited, and a large amount of manpower is required to implement monitoring. Summary of the invention

[0003] In order to overcome the above technical defects, the present invention provides an anti-theft method and system based on AI intelligence, which can improve the efficiency of monitoring work and reduce the manpower cost of monitoring work.

[0004] In order to solve the above problems, in a first aspect, the present invention discloses an anti-theft method based on AI intelligence, comprising the following steps:

[0005] Get surveillance video captured by the camera in real time;

[0006] Decode the surveillance video stream and perform AI intelligent analysis on the surveillance object in the surveillance video. AI intelligent analysis includes cross-camera tracking, human attribute recognition and / or suspicious action recognition;

[0007] If AI intelligent analysis finds that the monitored object has committed theft or damage, it will issue a warning to the monitored object on the spot and alert the staff in the monitoring room.

[0008] Furthermore, the video stream decoding of the surveillance video includes:

[0009] For each camera that shoots surveillance video, a corresponding processing process is created to decode the video stream.

[0010] Furthermore, the AI ​​intelligent analysis of the monitored object in the surveillance video includes cross-camera tracking, including:

[0011] When a monitored object is detected within the monitoring range, a tracking ID number is assigned to the monitored object, and the monitored object is tracked across cameras based on the tracking ID number.

[0012] Furthermore, the AI ​​intelligent analysis of the monitored object in the monitoring video includes human attribute recognition, including:

[0013] Identify the monitored object in the monitored video image frame after decoding the video stream to obtain feature parameters of the monitored object, wherein the feature parameters include age, gender and / or facial features;

[0014] Generate first portrait information of the monitored object based on the characteristic parameters;

[0015] Repeat the above steps to obtain a number of first portrait information, take the average characteristic value of the first portrait information as the second portrait information, and compare the second portrait information with the portrait information in the database.

[0016] Furthermore, the AI ​​intelligent analysis of the monitored object in the surveillance video includes suspicious action recognition, including:

[0017] Analyze whether the monitored object has any suspicious actions during the cross-camera tracking process, which include directly putting items into pockets and leaving the monitoring range with items on the body but not paying.

[0018] Furthermore, if the AI ​​intelligent analysis finds that the monitored object has been stolen, a warning will be issued to the monitored object on the spot and the staff in the monitoring room will be reminded to pay attention, including:

[0019] The second portrait information is successfully compared with the portrait information in the database, and the monitored object is judged to be a key monitored object, its monitoring weight is increased, and a voice prompt is issued to the monitored object on the spot.

[0020] Furthermore, if the AI ​​intelligent analysis finds that the monitored object has been stolen, a warning will be issued to the monitored object on the spot and the staff in the monitoring room will be reminded to pay attention, including:

[0021] If the monitored object has any suspicious behavior during the cross-camera tracking process, an alarm will be sent to the staff in the monitoring room, and the second portrait information of the monitored object will be stored in the database.

[0022] In a second aspect, the present invention further discloses an AI-based anti-theft system for implementing the above method, comprising:

[0023] A photography module, including several cameras, for shooting surveillance videos;

[0024] A video stream decoding module is connected to the camera module and is used to decode the video stream of the surveillance video, wherein a separate processing process is created for each camera to perform decoding;

[0025] The AI ​​algorithm module is connected to the video stream decoding module to analyze the decoded surveillance video, determine whether there is any theft or damage, and send corresponding warning information based on the analysis results.

[0026] Furthermore, it also includes:

[0027] The control platform is used to send control signals according to the staff's instructions, display the original surveillance video taken by the camera module, and display the theft and loss behavior results analyzed by the AI ​​algorithm module;

[0028] The monitoring module is connected to the control platform, the video stream decoding module and the AI ​​algorithm module respectively, and is used to forward the control signal sent by the control platform to the video stream decoding module and the AI ​​algorithm module, and to monitor the working status of the video stream decoding module and the AI ​​algorithm module.

[0029] Furthermore, the AI ​​algorithm module includes:

[0030] A ReID tracking algorithm unit is used to assign a tracking ID number to a monitored object when a monitored object is detected within the monitoring range, and to track the monitored object across cameras according to the tracking ID number;

[0031] A human attribute recognition unit is used to recognize the monitored object in the monitored video image frame after the video stream is decoded, obtain characteristic parameters of the monitored object, generate portrait information of the monitored object based on the characteristic parameters, and compare the portrait information with the portrait information in the database, wherein the characteristic parameters include age, gender and / or facial features;

[0032] The suspicious action recognition unit is used to analyze whether the monitored object has any suspicious actions during the cross-camera tracking process. Suspicious actions include directly putting items into pockets and carrying items on the body without paying when leaving the monitoring range.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The present invention discloses an anti-theft method and system based on AI intelligence, the anti-theft method comprising: obtaining surveillance videos shot by cameras in real time; decoding the surveillance videos, and performing AI intelligent analysis on the surveillance objects in the surveillance videos, the AI ​​intelligent analysis including cross-camera tracking, human attribute recognition and / or suspicious action recognition; if the AI ​​intelligent analysis finds that the surveillance object has theft or damage, a warning is issued to the surveillance object on the spot and the staff in the surveillance room are reminded to pay attention. By analyzing the surveillance video images with AI intelligence, theft or damage can be discovered in time, and the staff can be assisted in safety supervision, thereby improving work efficiency and reducing the labor cost of monitoring work. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The specific embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings, wherein:

[0036] Figure 1 This is a flow chart of the AI-based anti-theft method described in Example 1;

[0037] Figure 2 This is a schematic diagram of the structure of the AI-based anti-theft system described in Example 2;

[0038] Marking description: 100, photography module; 200, video stream decoding module; 300, AI algorithm module; 400, control platform; 500, monitoring module; 600, database. DETAILED DESCRIPTION

[0039] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0040] Example 1

[0041] This embodiment discloses an anti-theft method based on AI intelligence, such as Figure 1 , including the following steps:

[0042] S1. Obtain the surveillance video captured by the camera in real time.

[0043] S2. Decode the video stream of the surveillance video and perform AI intelligent analysis on the surveillance object in the surveillance video. The AI ​​intelligent analysis includes cross-camera tracking, human attribute recognition and / or suspicious action recognition.

[0044] Specifically, decoding the video stream of the surveillance video includes:

[0045] For each camera that shoots surveillance video, a corresponding processing process is created to decode the video stream, and configure the camera's resolution, focal length, viewing angle and other parameters. Each processing process does not interfere with each other. When a problem occurs in one processing process, other processing processes can decode normally.

[0046] In one embodiment, it also includes:

[0047] The surveillance video image frames after decoding the video stream are sent to the cloud for sharing.

[0048] In one embodiment, AI intelligent analysis is performed on the monitored object in the surveillance video, and the AI ​​intelligent analysis includes cross-camera tracking, including:

[0049] When a monitored object is detected within the monitoring range, a tracking ID number is assigned to the monitored object, and the monitored object is tracked across cameras according to the tracking ID number until the monitored object leaves the monitoring range.

[0050] To improve accuracy, the human shape and face of the monitored object are analyzed. First, yolov8 is used to detect the human shape and face in all cameras, and then EfficientNet is used to extract features and compare the human shape and face to complete the cross-camera pedestrian re-identification ReID function.

[0051] In one embodiment, AI intelligent analysis is performed on the monitored object in the monitoring video, and the AI ​​intelligent analysis includes human attribute recognition, including:

[0052] After the video stream is decoded, two lightweight resnet18 networks are used to identify the monitored objects in the surveillance video image frames tracked across cameras. Specifically, the detected human shapes and faces are analyzed for attributes to obtain the characteristic parameters of the monitored objects, where the characteristic parameters include age, gender and / or facial features;

[0053] Generate first portrait information of the monitored object based on the characteristic parameters;

[0054] Repeat the above steps to obtain a number of first portrait information, take the average characteristic value of the first portrait information as the second portrait information, and compare the second portrait information with the portrait information in the database.

[0055] In one embodiment, AI intelligent analysis is performed on the monitored object in the surveillance video, and the AI ​​intelligent analysis includes suspicious action recognition, including:

[0056] Analyze the behavior of the monitored object during cross-camera tracking, including walking, picking up goods, paying at checkout, putting goods into pockets, etc., and determine whether there are suspicious actions, among which suspicious actions include putting items directly into pockets in front of the shelf but not paying when leaving the monitoring range.

[0057] Yolov8 includes the detection of human skeleton key points. The 15 frames of continuously tracked human skeleton key points are input into the graph convolution network for feature extraction, and then input into the SVM support vector machine for suspicious action recognition and classification.

[0058] S3. If AI intelligent analysis finds that the monitored object has committed theft or damage, a warning will be issued to the monitored object on the spot and the staff in the monitoring room will be reminded to pay attention.

[0059] Specifically, if AI intelligent analysis finds that the monitored object has been stolen or damaged, it will issue a warning to the monitored object on the spot and alert the staff in the monitoring room, including:

[0060] If the second portrait information is successfully matched with the portrait information in the database, the monitored object is judged as a key monitored object, its monitoring weight is increased, and a voice prompt is given to the monitored object on the spot, where the monitoring weight is calculated as: conf=base+conf', where conf is the monitoring weight; base is the basic value, which defaults to 0. When the monitored object is a key monitored object, the basic value is increased; conf' is the theft and loss behavior value.

[0061] Specifically, if AI intelligent analysis finds that the monitored object has been stolen or damaged, it will issue a warning to the monitored object on the spot and alert the staff in the monitoring room, including:

[0062] If the monitored object has suspicious movements during the cross-camera tracking process, an alarm is sent to the staff in the monitoring room, and the second portrait information of the monitored object is stored in the database, and the theft behavior value of the second portrait information is increased.

[0063] Through AI intelligence, the surveillance video images can be analyzed to detect theft and loss in time, assisting staff in security supervision, improving work efficiency and reducing the labor cost and human resources of monitoring work. By storing the second portrait information of the monitored objects with suspicious actions in the database, it is convenient for subsequent analysis of key suspicious users. When these users are monitored again, they can be monitored and prompted, and provide a basis for security analysis and security arrangements of staff.

[0064] Example 2

[0065] Based on the same inventive concept, the embodiment of the present application also discloses an AI-based anti-theft system for implementing the AI-based anti-theft method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more AI-based anti-theft system embodiments provided below can refer to the limitations of the AI-based anti-theft method above, and will not be repeated here.

[0066] In one embodiment, an AI-based anti-theft system is provided. Figure 2 , including a photography module 100, a video stream decoding module 200, and an AI algorithm module 300; the photography module 100 includes several cameras for shooting surveillance videos; the video stream decoding module 200 is connected to the camera module 100, and is used to decode the video stream of the surveillance video, wherein a separate processing process is created for each camera for decoding; the AI ​​algorithm module 300 is connected to the video stream decoding module 200, and is used to analyze the decoded surveillance video, determine whether there is theft or damage, and send corresponding warning information according to the analysis results.

[0067] Specifically, the video stream decoding module 200 is also used to configure parameters such as resolution, focal length, and viewing angle for each camera.

[0068] In one embodiment, the video stream decoding module 200 sends the surveillance video image frames after decoding the video stream to the cloud for sharing so that the AI ​​algorithm module can perform the next step of analysis and processing.

[0069] Specifically, it also includes a control platform 400 and a monitoring module 500. The control platform is used to send control signals according to the instructions of the staff, display the original monitoring video taken by the photography module, and display the theft and damage results analyzed by the AI ​​algorithm module. The AI ​​algorithm module sends the alarm to the control platform through the RESTful API to prompt the staff; the AI ​​algorithm module also records the theft and damage event, records the screen recording ID number, and stores the screen recording file and the screen recording ID number in the local edge device database 600 for the staff to view and playback. The screen recording file also includes the store ID where the theft and damage event occurred, the address of the video stream, the schematic image frame and timestamp of the theft and damage event, etc. The staff confirms whether the theft and damage has occurred by watching the monitoring video and the theft and damage results displayed on the control platform. If the theft and damage is confirmed to have occurred, a voice prompt is issued to indicate that the monitored object is under surveillance.

[0070] The monitoring module 500 is a lightweight http server, which is connected to the control platform, video stream decoding module and AI algorithm module respectively through RESTful API, and is used to forward the control signal sent by the control platform 400 to the video stream decoding module 200 to start the corresponding processing process for video stream decoding, and forward it to the AI ​​algorithm module 300 to start algorithm analysis, and monitor the working status of the video stream decoding module 200 and the AI ​​algorithm module 300. Communication with the edge devices of the store, AI algorithm management and alarm reception are realized through the RESTful API.

[0071] Specifically, the AI ​​algorithm module 300 includes a ReID tracking algorithm unit, a human attribute recognition unit and a suspicious action recognition unit. The ReID tracking algorithm unit is used to assign a tracking ID number to the monitored object when a monitored object is detected within the monitoring range, and to track the monitored object across cameras based on the tracking ID number.

[0072] The human attribute recognition unit is used to identify the monitored object in the monitoring video image frame after the video stream is decoded, obtain the characteristic parameters of the monitored object, generate portrait information of the monitored object based on the characteristic parameters, and compare the portrait information with the portrait information in the database 600, wherein the characteristic parameters include age, gender and / or facial features.

[0073] The suspicious action recognition unit is used to analyze whether the monitored object has any suspicious actions during the cross-camera tracking process. Suspicious actions include directly putting items into pockets and carrying items on the body without paying when leaving the monitoring range.

[0074] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Therefore, any modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. An anti-theft method based on AI intelligence, characterized in that: The steps include: Get surveillance video captured by the camera in real time; Decode the surveillance video stream and perform AI intelligent analysis on the surveillance object in the surveillance video. AI intelligent analysis includes cross-camera tracking, human attribute recognition and / or suspicious action recognition; If AI intelligent analysis finds that the monitored object has committed theft or damage, it will issue a warning to the monitored object on the spot and alert the staff in the monitoring room.

2. The anti-theft method according to claim 1, characterized in that: The video stream decoding of the surveillance video includes: For each camera that shoots surveillance video, a corresponding processing process is created to decode the video stream.

3. The anti-theft method according to claim 1, characterized in that: The AI ​​intelligent analysis of the monitored object in the surveillance video includes cross-camera tracking, including: When a monitored object is detected within the monitoring range, a tracking ID number is assigned to the monitored object, and the monitored object is tracked across cameras based on the tracking ID number.

4. The anti-theft method according to claim 3, characterized in that: The AI ​​intelligent analysis of the monitored object in the monitoring video includes human attribute recognition, including: Identify the monitored object in the monitored video image frame after decoding the video stream to obtain feature parameters of the monitored object, wherein the feature parameters include age, gender and / or facial features; Generate first portrait information of the monitored object based on the characteristic parameters; Repeat the above steps to obtain a number of first portrait information, take the average characteristic value of the first portrait information as the second portrait information, and compare the second portrait information with the portrait information in the database.

5. The anti-theft method according to claim 4, characterized in that: The AI ​​intelligent analysis of the monitored object in the surveillance video includes suspicious action recognition, including: Analyze whether the monitored object has any suspicious actions during the cross-camera tracking process, which include directly putting items into pockets and leaving the monitoring range with items on the body but not paying.

6. The anti-theft method according to claim 4, characterized in that: If the AI ​​intelligent analysis finds that the monitored object has been stolen, it will issue a warning to the monitored object on the spot and remind the staff in the monitoring room to pay attention, including: The second portrait information is successfully compared with the portrait information in the database, and the monitored object is judged to be a key monitored object, its monitoring weight is increased, and a voice prompt is issued to the monitored object on the spot.

7. The anti-theft method according to claim 5, characterized in that: If the AI ​​intelligent analysis finds that the monitored object has been stolen, it will issue a warning to the monitored object on the spot and remind the staff in the monitoring room to pay attention, including: If the monitored object has any suspicious behavior during the cross-camera tracking process, an alarm will be sent to the staff in the monitoring room, and the second portrait information of the monitored object will be stored in the database.

8. An AI-based anti-theft system, used to implement any of the methods described in claims 1-7, characterized in that: include: A photography module, including several cameras, for shooting surveillance videos; A video stream decoding module, connected to the camera module, for decoding the surveillance video stream, wherein a separate processing process is created for each camera for decoding; The AI ​​algorithm module is connected to the video stream decoding module, and is used to analyze the decoded surveillance video, determine whether there is theft or damage, and send corresponding warning information based on the analysis results.

9. The anti-theft system according to claim 8, characterized in that: Also includes: The control platform is used to send control signals according to the staff's instructions, display the original surveillance video taken by the camera module, and display the theft and loss behavior results analyzed by the AI ​​algorithm module; The monitoring module is connected to the control platform, the video stream decoding module and the AI ​​algorithm module respectively, and is used to forward the control signal sent by the control platform to the video stream decoding module and the AI ​​algorithm module, and to monitor the working status of the video stream decoding module and the AI ​​algorithm module.

10. The anti-theft system according to claim 8, characterized in that: The AI ​​algorithm module includes: A ReID tracking algorithm unit is used to assign a tracking ID number to a monitored object when a monitored object is detected within the monitoring range, and to track the monitored object across cameras according to the tracking ID number; A human attribute recognition unit is used to recognize the monitored object in the monitored video image frame after the video stream is decoded, obtain characteristic parameters of the monitored object, generate portrait information of the monitored object based on the characteristic parameters, and compare the portrait information with the portrait information in the database, wherein the characteristic parameters include age, gender and / or facial features; The suspicious action recognition unit is used to analyze whether the monitored object has any suspicious actions during the cross-camera tracking process. Suspicious actions include directly putting items into pockets and carrying items on the body without paying when leaving the monitoring range.

Images