CN114241392A

CN114241392A - Automatic factory specification inspection method based on video behavior recognition

Info

Publication number: CN114241392A
Application number: CN202111593346.3A
Authority: CN
Inventors: 周凯凯; 高善恒
Original assignee: Suzhou Enterprise Intelligence Information Technology Co ltd
Current assignee: Suzhou Enterprise Intelligence Information Technology Co ltd
Priority date: 2021-12-23
Filing date: 2021-12-23
Publication date: 2022-03-25

Abstract

The invention discloses a factory specification automatic inspection method based on video behavior recognition. The method relates to the field of image recognition, in particular to an intelligent inspection method based on object behavior specification detection in a factory production scene environment. The behavior specification intelligent judgment and detection method only aims at scenes of violation detection, intelligent inspection and inspection problem processing in a factory production environment, and under the scenes, employees need to perform specific operation on equipment, products and production places according to unified regulations of enterprises in the production places. The method disclosed by the invention is used for intelligently judging whether the conditions of the production site, the production data arranged in the production site, the staff entering the production site and the staff linking all the production data and the produced products are normally operated according to the company definition specifications. If abnormal conditions occur, the intelligent routing inspection task is established, the rectification is informed, the rectification result is fed back intelligently, and finally the intelligent routing inspection of the production standard is realized.

Description

Automatic factory specification inspection method based on video behavior recognition

Technical Field

The invention relates to the technical field of computers, in particular to an algorithm application based on image recognition.

Background

Production management of a factory has been a complicated and troublesome problem. All plant managers want to be able to maintain safe production and efficient management of plants, and basically the more complex the plant production environment is, the more workers are, the more complex the products are produced, the larger the production shop is, the more blank the management is, the higher the management cost is and the greater the risk probability in the whole production process is. In the current background, almost all factories already have a good monitoring base, and basically, video coverage of a production management core area is realized. However, the current monitoring can only be used for post-investigation after problems occur, and for a factory manager, the significance of the overall safety value is small, and the management value is almost not provided, so that the resource waste is very large.

Under the background of basic perfect informatization, a production management party can generally know the basic conditions of 'employees', 'production data' and 'products' in three elements in the whole factory, and a manager can timely know the inventory condition and the production condition of the products in the factory. The more important 'staff' and 'production data' of the three elements are unknown, and most production behaviors are not obvious in the 'product' aspect.

At present, the factory inspection management process mainly depends on the manpower management of inspection personnel, and the inspection personnel are arranged at a specific time stage to inspect so as to ensure that all production processes are in compliance as much as possible. The whole process has the following disadvantages:

1. large consumption of manpower resources, long polling period and poor effect

2. The behavior standard depends on the judgment of manpower, and the uniform and standardized execution of enterprise regulations cannot be guaranteed

3. The consumption of manpower resources is huge, and the long-term complete coverage inspection cannot be realized

Disclosure of Invention

Aiming at the defects, the invention provides an intelligent, fair, efficient, unwieldy and extensible factory standard intelligent inspection method for the production places of the factories.

The technical scheme for solving the technical problems of the invention is as follows: a factory specification automatic inspection method based on video behavior recognition is disclosed. The method comprises continuous training of illegal behaviors in the production process of a factory, and related samples need to be continuously collected and marked in the training process, so that a better effect is achieved. In this process, an understanding of the various general and specific behavioral specifications, routine violations, and the like in the factory production process are included, and in the implementation, there is a need to continuously accumulate material, covering both correct paradigms and incorrect examples. The materials cover video materials and picture materials, and for the video materials, the video materials need to be imported into a GPU server, the materials are cut into pictures according to proper rules, and subsequent processing is carried out based on the pictures.

The picture processing described in the present invention includes behavior tagging and training of pictures. In this process, it is necessary to preset specific behaviors, and which performances or states of the specific behaviors are qualified and which performances and states are violations. For example, in a production site, a worker leaves the production site for a long time, behaviors obviously violate a behavior specification, scattered garbage or production product stacks violate a stacking specification in the production site, and pictures need to be marked, correct and wrong samples need to be marked, and the samples are input into a GPU server for behavior recognition training.

The training method and the training result of the invention can be written into corresponding equipment according to the requirements of different models. The object detection and basic behavior detection models can be written into the terminal equipment, object attribute and behavior detection in a basic production scene is completed by means of the edge computing capability of a terminal camera, basic data acquisition work is completed, and acquired data and video streams are transmitted to the central GPU server cluster together.

And after the central server acquires the edge calculation detection result and the corresponding picture, calculating the picture. And secondary detection is needed to be carried out on the monitoring behavior in the calculation, and the accuracy and authenticity of front-end detection are judged. Meanwhile, the acquisition camera can send acquisition pictures to the central processing server at regular time according to a certain time rule, and the central processing unit also needs to process the non-judgment pictures according to a set rule so as to detect whether other problems defined by enterprises exist.

The combination of the real-time pictures and the random pictures can greatly improve the flexibility and the accuracy of the illegal detection of the factory production behavior, and if a manager needs to check other information in the future, the manager can also process the pictures based on the illegal detection, thereby ensuring the expandability and the extensibility of the whole system.

After the central GPU processor cluster judges and classifies relevant identification problems of factory production scenes, the results are matched with pictures and transmitted to an intelligent inspection algorithm platform, and algorithm brands comprise a large number of algorithm models, such as a labeling algorithm for a production workshop, a labeling algorithm for staff, a prediction algorithm for possible problems and the like. After the algorithm platform processes the relevant problem, the problem is output. The output mode is to establish a set of 'automatic' intelligent inspection method for a factory manager. In the method, an inspection task is required to be established immediately, the inspection task is divided into 'environment violation', 'flow violation' and 'behavior violation', and the exceptions are issued regularly, for example, in the whole control center, the specific situation, the executor and the current problem of each task need to be determined, and the executor of the problem needs to acquire the problem, know the detailed situation of the problem, and whether the task is a sporadic task or a sudden task, so as to perform processing and confirmation. The intelligent system can continuously track the problems to ensure the execution of the problems, and meanwhile, the situation of the problems needs to be monitored, and the current changing progress of the established problems is intelligently judged. If the current problem has been completed, the intelligence changes the state of the task, thereby completing the fully intelligent closed loop of set-up-change-verification on the flow.

Drawings

The invention is further described below with reference to the accompanying drawings.

FIG. 1 is a schematic block diagram of a method for automatic plant specification inspection based on video behavior recognition.

Detailed Description

The invention is further described with reference to the following figures and specific examples.

The method is a factory standard automatic inspection method based on video behavior recognition, and based on the actual conditions of factory production and the actual conditions of large-scale production enterprise management departments, a proper camera is installed in a proper point location direction, an intelligent camera carrying an intelligent computing platform is selected as the camera, and the camera needs to cover the whole production management place. After the camera is erected, edge calculation is selected according to the requirements of production management to carry out face detection, face acquisition, scene detection and scene acquisition. The collected data are compared in scene specifications, the comparison conclusion data are transmitted to an intelligent inspection algorithm platform, based on the algorithm, abnormal information is filed according to abnormal reasons, inspection tasks are automatically established according to different types, and after the inspection tasks are automatically distributed to specific responsible personnel and rectification personnel. Meanwhile, the intelligent inspection algorithm platform can continuously track the abnormity according to the previous inspection task, and can automatically check and change the problem state and inform relevant personnel if relevant problems are tracked and corrected.

Example 1:

a technician selects a factory as an intelligent factory violation inspection case, and in the case, firstly, a proper camera is selected to be installed in a proper direction by selecting a proper point position according to the actual situation of a factory production place, the arrangement situation of production data and products in the factory. Generally, a workshop doorway corridor needs to be installed, the details of staff passing through a factory are known, and a production area is installed so as to know the production specification condition of the factory, install a core storage area and know the environment and safety condition of a storage area.

The technical personnel carry out configuration, the factory camera is accessed into the GPU server cluster, the face detection acquisition algorithm, the behavior recognition algorithm and the face recognition algorithm are deployed into the GPU server cluster, and the reliability of the image recognition algorithm is guaranteed. The GPU server firstly processes the accessed video, captures the monitored object, collects and judges the quality of the monitored object after detecting the human face or defining the behavior, and outputs an abnormal result to the intelligent inspection management system.

After the intelligent patrol detects the abnormality, the abnormal information is filed according to the abnormal reason, such as 'environmental violation', 'flow violation' and 'behavior violation'. And automatically establishing inspection tasks according to different types, and automatically distributing the tasks to specific responsible personnel and rectification personnel after the inspection tasks.

When the relevant manager receives the task in the system, the exception is rectified. In the rectification process, the intelligent patrol inspection system still commands the front-end camera to continuously track according to the previous patrol inspection task, and if relevant problems are tracked, the intelligent patrol inspection system automatically checks, changes the problem state and informs relevant personnel.

With the above embodiments, all plant managers can easily implement the present invention. Any technical engineer may implement the process in our way.

Claims

1. A factory specification automatic inspection method based on video behavior recognition is disclosed. The intelligent camera with a computing module is installed at a key production relevant position of a factory, local data transmission networks of various workshops and offices are erected, a centralized picture processing GPU cluster is arranged, and an information intelligent system for intelligent factory patrol based on image processing capacity is arranged. And at the data acquisition end, a face detection acquisition algorithm, a specific object detection acquisition algorithm and a specific behavior detection algorithm in a factory are written into the camera, so that the camera is based on edge calculation, and the technical capital investment cost of a large-scale factory is reduced. The edge computing system transmits the detected behaviors back to a centralized picture processing server set up by a factory management company in a picture or short video mode, deeply processes and analyzes the pictures, and transmits the analysis result to a factory management data center. Based on the data basis provided by data analysis of the data center, inspection tasks of different levels are automatically constructed according to the classification of detected problems and are issued to different responsible persons for processing. In the processing process, the intelligent inspection method can perform timed inspection and automatically check the effect of the intelligent inspection method for the problems which can be automatically identified intelligently, and finally complete the intelligent construction of the complete automatic inspection from intelligent problem discovery to intelligent tracking to intelligent checking.

2. The method as claimed in claim 1, wherein all workshop video capture devices in a factory or a group are connected to the management data center. The data acquisition camera simultaneously plays the role of acquiring and calculating the terminal in the data acquisition camera, so that the larger the coverage network of a production workshop and an office is, the stronger the calculation capacity is, meanwhile, the equipment transmits the behavior pictures and the short videos acquired by the terminal to a central processing server through a public network, and the central server deeply identifies and calculates the materials.

S1, data acquisition ends are dispersed in all production workshops and management offices, and meanwhile, the data acquisition ends are edge computing equipment and bear the computing part of a solution.

And S2, both edge calculation and center calculation can be used for detecting the face, detecting scenes, judging the quality of pictures and videos and intercepting proper materials.

And S3, based on the equipment information and the scene information, deep calculation and processing are performed in a centralized mode, so that the maximum release of the calculation force is ensured, and the sustainable optimization of the algorithm is ensured.

3. The intelligent inspection system according to claim 1, wherein the decision of the production site behavior is output to the intelligent inspection system, and the intelligent inspection system can judge and distribute various inspection tasks.

4. The method as claimed in claim 1, wherein after the problems in the factory production site are automatically detected and distributed, the problem correction condition is verified at regular time or randomly according to the problem type, so as to complete the full automatic operation of the whole intelligent system.