CN113989499B - Intelligent alarm method in bank scene based on artificial intelligence - Google Patents
Intelligent alarm method in bank scene based on artificial intelligence Download PDFInfo
- Publication number
- CN113989499B CN113989499B CN202111607199.0A CN202111607199A CN113989499B CN 113989499 B CN113989499 B CN 113989499B CN 202111607199 A CN202111607199 A CN 202111607199A CN 113989499 B CN113989499 B CN 113989499B
- Authority
- CN
- China
- Prior art keywords
- target
- bank
- scene
- customer
- target detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/02—Banking, e.g. interest calculation or account maintenance
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Finance (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Computing Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Technology Law (AREA)
- General Business, Economics & Management (AREA)
- Image Analysis (AREA)
- Alarm Systems (AREA)
Abstract
An intelligent alarm method based on artificial intelligence in bank scene belongs to the technical field of computer vision intelligent identification, and aims at judging the pertinence of the categories of employees and customers in a detected target. The method comprises the steps that people in a scene are divided into bank staff, security guards and customers according to occupation, and objects in the scene are divided into objects lost by the customers, mobile phones and billboards according to different types; meanwhile, an improved yolov5 target detection algorithm is provided, an ASFF self-adaptive feature fusion module is added into the original yolov5 target detection algorithm to detect and identify target objects in a bank scene, and the accuracy and the real-time performance of the algorithm on target detection and identification are improved.
Description
Technical Field
The invention discloses an intelligent alarm method based on artificial intelligence in a bank scene, and belongs to the technical field of computer vision intelligent identification.
Background
The bank is a financial institution which manages deposit, loan, exchange, savings and the like, undertakes credit mediation, and has the characteristics of diversified scale, numerous financial service devices, complicated access personnel, wide management and related range and the like. Therefore, under such a scenario, it often happens that the customer fails to receive efficient service due to passive idling of staff. In addition, although the customer does not have the intention of professional help, if the worker can be supervised to pay attention to the customer requirement, the customer service experience can be greatly improved, and the bank revenue generating amount is increased.
The monitoring method under the current bank scene mainly depends on the staff to artificially and uninterruptedly monitor whether the customer needs help or not and whether the staff behaviors are in compliance, safe and the like, and a large amount of manpower is needed. The traditional monitoring method mainly depends on a manual detection method, and the method has many defects, such as high labor intensity, low automation degree and the like, is easily influenced by subjective emotion of monitoring personnel, and has larger monitoring error due to the fact that the manpower is easily fatigued and the like.
Disclosure of Invention
Aiming at the defects of the prior art, the invention discloses an intelligent alarm method in a bank scene based on artificial intelligence. The invention relates to application of a deep learning algorithm in the field of computer vision, in particular to a method for automatically identifying position and category information of personnel and articles in a bank scene based on a deep neural network, and carrying out intelligent alarm according to a detection result.
The detailed technical scheme of the invention is as follows:
an intelligent alarm method in a bank scene based on artificial intelligence is characterized by comprising the following steps:
s1: collecting video streams under a normal working scene of a bank through a bank monitoring camera;
s2: acquiring the pictures of the bank scene at different moments in the video stream by a frame extraction method of the collected video stream so as to make a data set;
s3: then, performing off-line training on the data set by using a modified yolov 5-based target detection algorithm; the improved yolov 5-based target detection algorithm is as follows: adding an ASFF self-adaptive feature fusion module aiming at yolov5 target detection model, performing weighted fusion on the output features of the upper layer, and improving the detection effect of the algorithm, wherein the ASFF self-adaptive feature fusion module is used for performing weighted fusion on the output features of the upper layerPrinciple of the ASFF adaptive feature fusion module: input features to upper layers
Are respectively multiplied by weights
Formula (ii)
Is composed of
Said
After dimension reduction, the ranges are all made to be 0,1 through a softmax function]Internal and the sum is 1;
s4: the trained model detects and identifies people and objects marked in the scene, and meanwhile, the position and the class label of the people and the objects are obtained;
s5: and according to the detection result of the model, obtaining whether a customer needs help or not and/or whether a staff plays a mobile phone or not through a post-processing module.
According to a preferred embodiment of the present invention, the specific process of S1 is:
s11: deploying a monitoring camera under a bank scene;
s12: and respectively collecting video streams of videos shot by the monitoring camera in different time periods.
According to a preferred embodiment of the present invention, the specific process of S2 is:
s21: processing the collected video stream according to a frame extraction method, wherein the time interval is 2s, and extracting a picture from the video every 2 s;
s22: collecting all the collected pictures into a data set;
s23: labeling the data set by using an open source Labeling tool Labeling, Labeling people and objects appearing in the image, comprising: customer, customer missing package, phone, post, security and staff;
the personnel are classified into customer, security and staff; dividing objects into customer _ missing _ package, phone and poster according to different types;
s24: and (4) marking the marked data set according to the following steps of 8: 1: 1, and distributing the data sets out of order and dividing the data sets into a training set, a testing set and a verification set.
According to a preferred embodiment of the present invention, the specific process of S3 is:
on the data set marked in S24, training an improved yolov5 model to obtain an object detection algorithm for identifying the position and the category information of people and articles in a bank scene based on improved yolov 5.
According to a preferred embodiment of the present invention, the specific process of S4 is:
s41: applying the target detection algorithm obtained in the step S3 to the data image in the bank scene obtained in the step S2, detecting the personnel and the articles in the image, and forming a target detection frame;
s42: and counting the detection results of the same frame according to the detection result of the S41, and displaying the result in the video frame.
According to a preferred embodiment of the present invention, the specific process of S5 is:
s51: and (3) performing circular traversal on all detected targets, and firstly judging whether the target is an employee:
if the mobile phone is the employee, judging whether the mobile phone exists in the target detection frame intersected by the other target detection frame and the employee detection frame, if so, giving an alarm to a background: the staff plays the mobile phone;
if the target is judged not to be the staff, whether the target is the customer is judged, if so, whether the target intersected with the other target detection boxes by the customer detection boxes has the staff exists is judged, and if not, a background is alarmed: there is a customer need for assistance;
s52: if the target is neither employee nor customer, the target is skipped and the post-processing module ends after all targets have been traversed.
Compared with the prior art, the invention has the following beneficial effects:
1) the invention designs an intelligent alarm method in a bank scene based on artificial intelligence, and provides a new labeling category method, which comprises the steps of dividing characters in the scene into bank staff, security guards and customers according to occupation, dividing articles in the scene into articles lost by the customers, mobile phones and billboards according to different types.
2) The invention designs an intelligent alarm method based on artificial intelligence in a bank scene, provides an improved yolov5 target detection algorithm, adds an ASFF self-adaptive feature fusion module to an original yolov5 target detection algorithm, detects and identifies target objects in the bank scene, and improves the accuracy and the real-time performance of the algorithm on the target detection and identification.
3) The invention designs a post-processing method based on a model detection result, provides a new logic judgment method, carries out targeted judgment on the categories of employees and customers in a detected target, designs different logics according to the specificity of the target, and improves the intelligence degree of an alarm system.
Drawings
FIG. 1 is a flow chart of the intelligent alarm method of the invention.
FIG. 2 is a labeled block diagram of a data set in a banking scenario.
Fig. 3 is a schematic diagram of ASFF adaptive feature fusion.
FIG. 4 is a schematic diagram of a scene in which the recognition result of the target detection model is alarm-free.
FIG. 5 is a schematic diagram of a scenario in which the object detection model identifies that an employee plays a mobile phone alarm.
FIG. 6 is a diagram of a scenario in which an object detection model identifies that a customer needs help to alert.
Detailed Description
The present invention is further illustrated by, but not limited to, the following examples.
Examples of the following,
As shown in fig. 1, an intelligent alarm method in a bank scene based on artificial intelligence includes:
s1: collecting video streams under a normal working scene of a bank through a bank monitoring camera;
s2: acquiring the pictures of the bank scene at different moments in the video stream by a frame extraction method of the collected video stream so as to make a data set;
s3: then, performing off-line training on the data set by using a modified yolov 5-based target detection algorithm; the improved yolov 5-based target detection algorithm is as follows: adding an ASFF adaptive feature fusion module aiming at a yolov5 target detection model, performing weighted fusion on the output features of an upper layer, and improving the detection effect of an algorithm, wherein the principle of the ASFF adaptive feature fusion module is as follows: input features to upper layers
Are respectively multiplied by weights
Formula (ii)
Is composed of
Said
After dimension reduction, the ranges are all made to be 0,1 through a softmax function]Internal and the sum is 1;
s4: the trained model detects and identifies people and objects marked in the scene, and meanwhile, the position and the class label of the people and the objects are obtained;
s5: and according to the detection result of the model, obtaining whether a customer needs help or not and/or whether a staff plays a mobile phone or not through a post-processing module.
The specific process of S1 is as follows:
s11: deploying a monitoring camera under a bank scene;
s12: and respectively collecting video streams of videos shot by the monitoring camera in different time periods.
The specific process of S2 is as follows:
s21: processing the collected video stream according to a frame extraction method, wherein the time interval is 2s, and extracting a picture from the video every 2 s;
s22: collecting all the collected pictures into a data set;
s23: labeling the data set by using an open source Labeling tool Labeling, Labeling people and objects appearing in the image, comprising: customer, customer missing package, phone, post, security and staff;
s24: and (4) marking the marked data set according to the following steps of 8: 1: 1, and distributing the data sets out of order and dividing the data sets into a training set, a testing set and a verification set.
The specific process of S3 is as follows:
on the data set marked in S24, training an improved yolov5 model to obtain an object detection algorithm for identifying the position and the category information of people and articles in a bank scene based on improved yolov 5.
The specific process of S4 is as follows:
s41: applying the target detection algorithm obtained in the step S3 to the data image in the bank scene obtained in the step S2, detecting the personnel and the articles in the image, and forming a target detection frame;
s42: and counting the detection results of the same frame according to the detection result of the S41, and displaying the result in the video frame.
The specific process of S5 is as follows:
s51: and (3) performing circular traversal on all detected targets, and firstly judging whether the target is an employee:
if the mobile phone is the employee, judging whether the mobile phone exists in the target detection frame intersected by the other target detection frame and the employee detection frame, if so, giving an alarm to a background: the staff plays the mobile phone;
if the target is judged not to be the staff, whether the target is the customer is judged, if so, whether the target intersected with the other target detection boxes by the customer detection boxes has the staff exists is judged, and if not, a background is alarmed: there is a customer need for assistance;
s52: if the target is neither employee nor customer, the target is skipped and the post-processing module ends after all targets have been traversed.
Application examples,
And a certain banking point monitors real-time shooting, extracts pictures from the video stream and sends the pictures into the target detection model, and then detects target information at a certain moment. The data of the detection results of the staff and the customers are analyzed, whether the staff plays the mobile phone or not and whether the customers need help or not are judged according to the results, and the alarm is given to the background management personnel, so that the bank personnel and the business management are facilitated. The intelligent alarm under the bank scene comprises the following specific steps:
s1: collecting shooting data in a bank scene, wherein 13846 pictures are collected, training an improved yolov5 model, and referring to fig. 2, marking the data pictures, and fig. 3 is an ASFF adaptive feature fusion schematic diagram;
s2: acquiring detection result image data in a bank scene according to the identification result of the target detection model;
s3: performing cyclic traversal on all detected targets, firstly judging whether the target is a staff or not, if so, judging whether a mobile phone exists in the targets intersected with the staff detection boxes of other target detection boxes or not, if so, giving an alarm to a background, and allowing the staff to play the mobile phone, wherein in the attached figure 5, the situation that the staff plays the mobile phone is detected, and warning information is given;
s4: if the target is not the staff, judging whether the target is the customer, if so, judging whether the target intersected with the other target detection boxes by the customer detection boxes has staff, if not, giving an alarm to a background and helping the customer, and if so, detecting that the customer needs help and giving out warning information according to the attached figure 6;
s5: if the target is neither employee nor customer, the target is skipped, FIG. 4 is a normal case without alarm information, and the post-processing module is ended after all targets have been traversed.
In the invention, the recognition result of the image data in the bank scene is obtained through an improved yolov5 target detection algorithm, the result is analyzed, if the conditions that the employee plays the mobile phone and the customer needs help exist, corresponding alarm information is given, the high-efficiency operation of the bank work can be accurately ensured in real time, and the service is better provided for the customer.
Claims (1)
1. An intelligent alarm method in a bank scene based on artificial intelligence is characterized by comprising the following steps:
s1: collecting video streams under a normal working scene of a bank through a bank monitoring camera;
s2: acquiring the pictures of the bank scene at different moments in the video stream by a frame extraction method of the collected video stream so as to make a data set;
s3: then, performing off-line training on the data set by using a modified yolov 5-based target detection algorithm; the improved yolov 5-based target detection algorithm is as follows: adding an ASFF adaptive feature fusion module aiming at a yolov5 target detection model, and performing weighted fusion on the output features of the upper layer, wherein the principle of the ASFF adaptive feature fusion module is as follows: input features to upper layers
Are respectively multiplied by weights
Is of the formula
Said
After dimension reduction, the ranges are all made to be 0,1 through a softmax function]Internal and the sum is 1;
s4: the trained model detects and identifies people and objects marked in the scene, and meanwhile, the position and the class label of the people and the objects are obtained;
s5: according to the detection result of the model, whether a customer needs help or not and/or whether a staff plays a mobile phone or not are obtained through a post-processing module;
the specific process of S1 is as follows:
s11: deploying a monitoring camera under a bank scene;
s12: respectively collecting video streams of videos shot by a monitoring camera in different time periods;
the specific process of S2 is as follows:
s21: processing the collected video stream according to a frame extraction method, wherein the time interval is 2s, and extracting a picture from the video every 2 s;
s22: collecting all the collected pictures into a data set;
s23: labeling the data set by using an open source Labeling tool Labeling, Labeling people and objects appearing in the image, comprising: customer, customer _ missing _ package customer lost goods, phone handset, post billboard, securer security and staff bank clerk;
s24: and (4) marking the marked data set according to the following steps of 8: 1: 1, divided into a training set, a test set and a verification set;
the specific process of S3 is as follows:
training an improved yolov5 model on the data set marked in S24 to obtain a target detection algorithm for identifying the position and the category information of people and articles in a bank scene based on improved yolov 5;
the specific process of S4 is as follows:
s41: applying the target detection algorithm obtained in the step S3 to the data image in the bank scene obtained in the step S2, detecting the personnel and the articles in the image, and forming a target detection frame;
s42: counting the detection result of the same frame according to the detection result of S41, and displaying the result in the video frame;
the specific process of S5 is as follows:
s51: and (3) performing circular traversal on all detected targets, and firstly judging whether the target is an employee:
if the mobile phone is the employee, judging whether the mobile phone exists in the target detection frame intersected by the other target detection frame and the employee detection frame, if so, giving an alarm to a background: the staff plays the mobile phone;
if the target is judged not to be the staff, whether the target is the customer is judged, if so, whether the target intersected with the other target detection boxes by the customer detection boxes has the staff exists is judged, and if not, a background is alarmed: there is a customer need for assistance;
s52: if the target is neither employee nor customer, the target is skipped and the post-processing module ends after all targets have been traversed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111607199.0A CN113989499B (en) | 2021-12-27 | 2021-12-27 | Intelligent alarm method in bank scene based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111607199.0A CN113989499B (en) | 2021-12-27 | 2021-12-27 | Intelligent alarm method in bank scene based on artificial intelligence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113989499A CN113989499A (en) | 2022-01-28 |
| CN113989499B true CN113989499B (en) | 2022-03-29 |
Family
ID=79734373
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111607199.0A Active CN113989499B (en) | 2021-12-27 | 2021-12-27 | Intelligent alarm method in bank scene based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113989499B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115222946B (en) * | 2022-09-19 | 2022-11-25 | 南京信息工程大学 | Single-stage instance image segmentation method and device and computer equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086971A (en) * | 2018-07-09 | 2018-12-25 | 河南深实科技有限公司 | A kind of salesman's dispatch system based on customer's Activity recognition in solid shop/brick and mortar store |
| CN109948490A (en) * | 2019-03-11 | 2019-06-28 | 浙江工业大学 | A kind of employee's specific behavior recording method identified again based on pedestrian |
| CN111079694A (en) * | 2019-12-28 | 2020-04-28 | 神思电子技术股份有限公司 | Counter assistant job function monitoring device and method |
| WO2020222236A1 (en) * | 2019-04-30 | 2020-11-05 | Tracxone Ltd | System and methods for customer action verification in a shopping cart and point of sale |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113269039A (en) * | 2021-04-21 | 2021-08-17 | 南京鸣赫信息技术有限公司 | On-duty personnel behavior identification method and system |
-
2021
- 2021-12-27 CN CN202111607199.0A patent/CN113989499B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086971A (en) * | 2018-07-09 | 2018-12-25 | 河南深实科技有限公司 | A kind of salesman's dispatch system based on customer's Activity recognition in solid shop/brick and mortar store |
| CN109948490A (en) * | 2019-03-11 | 2019-06-28 | 浙江工业大学 | A kind of employee's specific behavior recording method identified again based on pedestrian |
| WO2020222236A1 (en) * | 2019-04-30 | 2020-11-05 | Tracxone Ltd | System and methods for customer action verification in a shopping cart and point of sale |
| CN111079694A (en) * | 2019-12-28 | 2020-04-28 | 神思电子技术股份有限公司 | Counter assistant job function monitoring device and method |
Non-Patent Citations (2)
| Title |
|---|
| Learning Spatial Fusion for Single-Shot Object Detection;Songtao Liu 等;《arXiv》;20191130;第1-10页 * |
| 基于改进YOLO v3的相似外部特征人员检测算法;梁思源等;《平顶山学院学报》;20200425(第02期);第52-59页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113989499A (en) | 2022-01-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109271554B (en) | Intelligent video identification system and application thereof | |
| CN106355154B (en) | Method for detecting frequent passing of people in surveillance video | |
| CN110728252A (en) | Face detection method applied to regional personnel motion trail monitoring | |
| Aljarrah et al. | Video content analysis using convolutional neural networks | |
| CN113989499B (en) | Intelligent alarm method in bank scene based on artificial intelligence | |
| CN112861673A (en) | False alarm removal early warning method and system for multi-target detection of surveillance video | |
| CN110458198A (en) | Multiresolution target identification method and device | |
| WO2023071188A1 (en) | Abnormal-behavior detection method and apparatus, and electronic device and storage medium | |
| Liu et al. | Video content analysis for compliance audit in finance and security industry | |
| CN107944373A (en) | A kind of video anomaly detection method based on deep learning | |
| CN111199172A (en) | Terminal screen recording-based processing method and device and storage medium | |
| CN114360182B (en) | Intelligent alarm method, device, equipment and storage medium | |
| CN109345427A (en) | The classroom video point of a kind of combination recognition of face and pedestrian's identification technology is to method | |
| KR101513180B1 (en) | System for a real-time cashing event summarization in surveillance images and the method thereof | |
| CN210091231U (en) | Wisdom garden management system | |
| CN115859689B (en) | Panoramic visualization digital twin application method | |
| CN112417989A (en) | Invigilator violation identification method and system | |
| CN116419059A (en) | Automatic monitoring method, device, equipment and medium based on behavior label | |
| Baliniskite et al. | Affective state based anomaly detection in crowd | |
| CN110533889A (en) | A kind of sensitizing range electronic equipment monitoring positioning device and method | |
| CN110309737A (en) | A kind of information processing method applied to cigarette sales counter, apparatus and system | |
| CN111191498A (en) | Behavior recognition method and related product | |
| Zhou et al. | Anomalous event detection based on self-organizing map for supermarket monitoring | |
| CN114170548A (en) | Oil field on-site micro-target detection method and system based on deep learning | |
| CN112766118A (en) | Object identification method, device, electronic equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |