KR102466433B1 - Device and method for recognizing work motion based on image analysis - Google Patents

Abstract

An apparatus for recognizing work motion based on image analysis according to an embodiment of the present invention includes a work image collection unit for obtaining image data about a work site captured through a camera; and a work motion recognition unit extracting feature information for each frame from the image data and recognizing a work motion of a worker in the work site by analyzing the extracted feature information for each frame.

Description

Apparatus and method for recognizing work motion based on image analysis {DEVICE AND METHOD FOR RECOGNIZING WORK MOTION BASED ON IMAGE ANALYSIS}

Embodiments of the present invention relate to a technology capable of monitoring the work operation of a worker at a work site, and more particularly, based on analysis of image data about the work site taken by a camera, the work of a worker in the work site An apparatus and method for recognizing work motions based on video analysis capable of realizing optimization of work motions through work monitoring, such as recognizing unnecessary motions by recognizing motions, and the like.

In the case of general production process management, production process management is performed using calculated values by setting the amount of production per hour or per day for each worker and converting the amount produced after a certain period of time into data. In fact, in the case of a general manufacturing company, work efficiency measurement for each worker takes a method of measuring work efficiency by determining the standard production volume that one person should produce in one hour and comparing the actual production value after a certain period of time.

The method of measuring work efficiency is generally calculated in the form of 'work efficiency (%) = actual production / standard production amount * 100' as follows, and the data generated at this time is documented or made into a database, and then the production process is managed using the data analyzed. It is common to apply to In this case, after a certain period of time, the production process analysis data depends only on the numerical data on the product output per hour of each worker for each process, and the production process analysis is also calculated by the supervisor or the person in charge An analysis of the production process is performed.

At this time, in the analysis of the production process where the problem is judged to have occurred, what kind of problems have occurred other than the quantity of production that cause the difference in production at each stage, and what kind of problem causes the work efficiency of each worker to differ from that of other workers in the same work process. Data on detailed production process management for etc. are lacking.

In order to solve this problem, a supervisor or human resource to monitor the production process is required for detailed analysis of the production process that is judged to have a problem, which causes unnecessary waste of human resources and increases the manufacturing cost of the product. As a factor, it causes a hindrance to product cost reduction.

As related prior art, there is Korean Patent Publication No. 10-2009-0131610 (Title of Invention: Production Process Management Method, Publication Date: 2009.12.29.).

An embodiment of the present invention realizes optimization of work motion through work monitoring, such as recognizing unnecessary motions by recognizing the work motions of workers in the work site based on the analysis of video data related to the work site captured by the camera. An image analysis-based task motion recognition device and method are provided.

The problem to be solved by the present invention is not limited to the above-mentioned problem (s), and another problem (s) not mentioned will be clearly understood by those skilled in the art from the following description.

An apparatus for recognizing work motion based on image analysis according to an embodiment of the present invention includes a work image collection unit for obtaining image data about a work site captured through a camera; and a work motion recognition unit extracting feature information for each frame from the image data and recognizing a work motion of a worker in the work site by analyzing the extracted feature information for each frame.

The work motion recognition unit includes at least one of a joint angle, a joint position, a torso position, a hand position, and a line of sight from the image data using at least one technique of joint recognition, hand position recognition, and semantic segmentation. It may include a feature information extraction module that extracts feature information.

The feature information extraction module may extract the feature information by further using a technique for recognizing at least one of motion and sound of the device when the operator works using the device.

The work action recognition unit analyzes the feature information for each frame to classify work actions constituting one process into semantic units, and subdivides the work actions from the feature information for each frame based on the work actions classified into each semantic unit. and recognizes it, and may include an action sequence analysis module for deriving the order of the subdivided work actions.

The operation sequence analysis module may include a classifier for determining the subdivided work motions and sequentially stacking each subdivided work motion to generate a work action sequence.

The classifier is derived by learning a deep learning model through data of a learning data library, and the learning data library matches detailed motion names, detailed motion task images, and feature information for each frame as data related to the subdivided task motion. It can be a library to store.

The classifier processes the input feature information for each frame to generate an output detailed action name, extracts a detailed action name matching the input feature information for each frame from the learning data library, and compares it with the output detailed action name; If a loss occurs according to the result of the comparison, a weight update may be performed.

The classifier may be derived by identifying a rule-based relationship between the feature information for each frame and the subdivided work motion.

An apparatus for recognizing work motions based on image analysis according to an embodiment of the present invention performs labeling for each video section based on the sequence of the subdivided work motions, classifies the subdivided work motions for each video section through the labeling, and It may further include a task management unit that outputs output on the screen of the monitoring terminal.

An image analysis-based work motion recognition method according to an embodiment of the present invention includes the steps of obtaining image data about a work site captured by a working image collection unit of the image analysis-based work motion recognition device through a camera; extracting feature information for each frame from the image data by a work motion recognition unit of the image analysis-based work motion recognition device; and recognizing, by the work motion recognition unit, a work motion of a worker in the work site by analyzing the extracted feature information for each frame.

The step of extracting feature information for each frame from the image data may include joint angles, joint positions, body positions, and hand positions from the image data using at least one technique of joint recognition, hand position recognition, and semantic segmentation. , extracting feature information including at least one of the line of sight.

Recognizing the work motion may include analyzing the feature information for each frame and classifying the work motion constituting one process into semantic units; segmenting and recognizing the work motion from the feature information for each frame based on the work motions classified for each semantic unit; and deriving an order of the subdivided work operations.

Details of other embodiments are included in the detailed description and accompanying drawings.

According to an embodiment of the present invention, based on the analysis of video data on the work site captured by a camera, work motion optimization through work monitoring such as recognizing unnecessary motions by recognizing the work motions of workers in the work site can be realized.

According to an embodiment of the present invention, since it is possible to recognize/display which operation is performed for each video section, it can be used for content creation for education and training.

1 is a block diagram illustrating an apparatus for recognizing an image analysis-based work motion according to an embodiment of the present invention.
FIG. 2 is a block diagram illustrating a detailed configuration of a work motion recognition unit of FIG. 1 .
Figure 3 is an exemplary view shown to show the installation position of the camera.
4 and 5 are diagrams illustrating an example of extracting feature information for each image frame.
6 and 7 are diagrams illustrating an example of subdividing a work motion through a classifier and generating a subdivided work motion sequence.
8 is a diagram showing an example of an output of a result of analyzing a single operator.
9 is a diagram illustrating an example of an output of a result of individually analyzing several operators.
10 is a flowchart illustrating a method for recognizing a work motion based on image analysis according to an embodiment of the present invention.

Advantages and/or features of the present invention, and methods of achieving them, will become apparent with reference to the detailed description of the following embodiments taken in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and will be implemented in various forms different from each other, only these embodiments make the disclosure of the present invention complete, and common knowledge in the art to which the present invention pertains. It is provided to completely inform the person who has the scope of the invention, and the present invention is only defined by the scope of the claims. Like reference numbers designate like elements throughout the specification.

In addition, preferred embodiments of the present invention to be carried out below are provided in each system functional configuration in order to efficiently explain the technical components constituting the present invention, or system functions commonly provided in the technical field to which the present invention belongs. The configuration is omitted as much as possible, and the functional configuration that should be additionally provided for the present invention will be mainly described. If one of ordinary skill in the art to which the present invention pertains, one will be able to easily understand the functions of conventionally used components among the omitted functional configurations not shown below, and also the omitted configurations as described above. The relationship between elements and components added for the present invention will also be clearly understood.

In addition, in the following description, "transmission", "communication", "transmission", "reception" and other similar terms of signals or information refer to direct transmission of signals or information from one component to another. as well as passing through other components. In particular, "transmitting" or "transmitting" a signal or information as a component indicates the final destination of the signal or information, and does not mean a direct destination. The same is true for "reception" of signals or information.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram illustrating an apparatus for recognizing an image analysis-based work motion according to an embodiment of the present invention.

Referring to FIG. 1 , an image analysis-based work motion recognition apparatus 100 according to an embodiment of the present invention includes a work image collection unit 110, a work motion recognition unit 120, a job management unit 130, and a control unit ( 140) may be included.

The work image collection unit 110 may obtain image data about a work site photographed through a camera. Here, the camera is installed in front of the worker as shown in (a) of FIG. 3 so as to capture the work scene at all times, and is installed on top of the workers as shown in (b) of FIG. 3 to capture the work scene of all workers. You can shoot at any time.

In this way, the camera may be installed in various positions such as the ceiling, the front of the work table, and the side of the work table to take images of work at various angles. On the other hand, the camera may be configured to include a depth camera (Depth Camera) to improve the precision of image analysis.

The work motion recognition unit 120 may recognize the work motion of a worker in the work site by analyzing the image data. That is, the work motion recognition unit 120 may extract feature information for each frame from the image data and analyze the extracted feature information for each frame to recognize the work motion of a worker in the work site.

To this end, the work action recognition unit 120 may include a feature information extraction module 210 and an action sequence analysis module 220 as shown in FIG. 2 . For reference, FIG. 2 is a block diagram illustrating a detailed configuration of the work motion recognition unit 120 of FIG. 1 .

The feature information extraction module 210 uses at least one of joint recognition, hand position recognition, and semantic segmentation, and at least one of joint angle, joint position, torso position, hand position, and line of sight from the image data. Feature information including one may be extracted. In this case, the feature information extraction module 210 may extract the feature information by further using a technique for recognizing at least one of motion and sound of the device when the operator works using the device.

That is, the feature information extraction module 210 can extract meaningful feature information for each image frame by continuously repeating a process of extracting meaningful feature information from one working image frame. This can be said to be a process of reducing high-dimensional image data to meaningful low-dimensional data.

For example, in the case of a working image frame as in the example of FIG. 4 , the feature information extraction module 210 may extract feature information from the working image frame through joint recognition, hand position recognition, and the like.

At this time, as shown in FIG. 5, the feature information extraction module 210 is applied to each frame of the working video frame over time, such as joint angle, joint position, body position, hand position, line of sight, etc. for each video frame. The feature information of can be extracted. Accordingly, the feature information extraction module 210 extracts different feature information for each frame.

The operation sequence analysis module 220 may analyze the feature information for each frame extracted by the feature information extraction module 210 and classify work operations constituting one process by semantic unit based on the analysis result.

The operation sequence analysis module 220 may segment and recognize the work operation from the feature information for each frame based on the work operations classified according to the semantic unit. Also, the operation sequence analysis module 220 may derive the sequence of the subdivided work operations.

To this end, the action sequence analysis module 220 may include a classifier that determines the subdivided work motions and builds each subdivided work motion in order to generate a work motion sequence.

That is, the action sequence analysis module 220 may determine the subdivided work actions through the classifier, and may generate a work action sequence by sequentially stacking the derived work actions.

For example, as shown in FIG. 6, the operation sequence analysis module 220 receives feature information for each image frame from the feature information extraction module 210 and 'gets' a work operation through the classifier. Recognition can be subdivided into 'transfer operation', 'loop assembly operation', 'pattern sewing operation', and 'next transfer operation'.

In addition, the action sequence analysis module 220 stacks the subdivided work actions in order according to the order of shooting time, and generates a work action sequence in the order of 'import', 'hook assembly', 'pattern sewing', and next delivery'. can do.

Here, the classifier may be built based on learning or rules. When built based on the learning, as shown in FIG. 7 , the classifier may be derived by learning a deep learning model through data of a learning data library. In this case, the classifier may utilize a model structure such as a convolutional neural network (CNN), a recurrent neural network (RNN), or a long short-term memory (LSTM).

Here, the learning data library is a library that matches detailed motion names, detailed motion work images, and feature information for each frame and stores them as data related to the subdivided work motions.

Referring continuously to FIG. 7, in the case of the classifier built on the learning basis, the input feature information for each frame is processed to generate an output detailed operation name, and matching with the input feature information for each frame from the learning data library. A detailed operation name corresponding to the output is extracted and compared with the output detailed operation name, and if a loss occurs according to the result of the comparison, a weight update may be performed.

Thus, according to an embodiment of the present invention, the subdivided work motions and their order can be more accurately classified and recognized by continuously learning and updating the deep learning model through the data of the learning data library, and through this, the work process Not only can optimization of work operations be realized through work monitoring, such as identifying unnecessary movements, but also it can be used for content production for education and training because it can recognize/display what work is being performed for each video section.

When built based on the rule, the classifier may be derived by identifying the relationship between the feature information for each frame and the subdivided work action based on a rule.

Meanwhile, the work motion recognition unit 120 may analyze the image data by individually applying an image processing algorithm based on at least one of a motion, a work object, and a region or applying two or more in combination.

Here, the image processing algorithm based on the motion recognizes and tracks the movement of the worker's joints or the worker itself, or the movement of a specific part of the worker's body (head, hand, etc.) or the decoration worn by the worker (uniform, wristband, etc.) Movement can be recognized and tracked.

As an embodiment, when the motion-based image processing algorithm is applied, the work motion recognition unit 120 recognizes the motion of the operator's joints or the motion of the operator itself from the image data, tracks the recognized motion, and It is possible to recognize the type of operation of the worker's work.

An image processing algorithm based on the workpiece may recognize and track the workpiece worn by the operator and determine the flow characteristics of the workpiece.

As another embodiment, when the workpiece-based image processing algorithm is applied, the work action recognition unit 120 recognizes the workpiece worn by the worker from the image data and recognizes the recognized work. By tracking the water and grasping its movement flow characteristics, it is possible to recognize the type of work motion of the worker.

An image processing algorithm based on the region may detect a change of a specific object in a specific region and output a change value.

As another embodiment, when the region-based image processing algorithm is applied, the work motion recognition unit 120 extracts a specific region from the image data, and in the extracted specific region, a specific object, for example, a job. It is possible to output the change value of the specific object by recognizing and tracking the ruler or the workpiece and detecting the change thereof, and through this, it is possible to recognize what kind of work action the corresponding operator is.

The work action recognition unit 120 may transmit the result (subdivision/sequence) of the work action to a monitoring terminal (not shown) through the work manager 130 described below.

The task management unit 130 may output task management data including a result of the task operation to the screen of the monitoring terminal. At this time, the task management unit 130 performs labeling for each video section based on the sequence of the subdivided work operations, classifies the subdivided work operations for each video section through the labeling, and outputs the subdivided work operations to the screen of the monitoring terminal. can do.

For example, as shown in FIG. 8 , when the work management unit 130 recognizes a single worker's work motion according to the analysis result of the image data, the work manager 130 arranges and analyzes the single worker's subdivided work motion according to the work order. It can be output as a result.

At this time, the task management unit 130 may include the current time, instrument unique ID value, and required task time in the task management data whenever a task is completed, and transmit the same to the monitoring terminal. Accordingly, the monitoring terminal may receive the task management data and display the received task management data on a screen.

On the other hand, as shown in FIG. 9 , when the work motions of several workers are recognized according to the analysis result of the image data, the work manager 130 arranges and analyzes the work motions subdivided for each worker according to the work order. It can be output as a result.

That is, in the case of worker 1, the work management unit 130 converts work actions in the order of 'import', 'hook assembly', 'pattern sewing', and 'next delivery' as a result of analyzing the subdivided work motions of worker 1. In the case of worker 2, the work actions in the order of 'import', 'hook assembly', 'pattern sewing', and 'next delivery' are output as an analysis result of the subdivided work actions of worker 2, In the case of 3, work actions consisting of 'import', 'glue', 'hook assembly', 'pattern sewing', and 'next delivery' in the order can be output as the analysis result of the subdivided work motion of the worker 3.

At this time, the task management unit 140 may include the current time, instrument-specific ID value, operator-specific ID value, and required task time in the task management data whenever a task is completed for each worker, and transmit the data to the monitoring terminal. Accordingly, the monitoring terminal may receive the task management data and display the received task management data on a screen.

The control unit 150 is an image analysis-based work motion recognition device 100 according to an embodiment of the present invention, that is, the work image collection unit 110, the work motion recognizer 120, and the work management unit 130 You can control the overall operation of the etc.

Meanwhile, the image analysis-based work motion recognition apparatus 100 according to an embodiment of the present invention processes a process of photographing and analyzing a work image and transmitting the analysis result (subdivided work motion and its sequence). A distributed processing method or a central processing method may be used for this purpose.

In the case of using the distributed processing method, the image analysis-based task motion recognition device 100 according to an embodiment of the present invention may be implemented in a form included in a measuring instrument (not shown), and the image analysis result is transmitted to a gateway It can be transmitted to a production management system (not shown) through. The production management system may include the aforementioned monitoring terminal.

In contrast, in the case of using the central processing method, the image analysis-based task motion recognition device 100 according to an embodiment of the present invention may be implemented in a form included in an analysis server (not shown), An image may be received and analyzed through a gateway, and the analysis result may be transmitted to the production management system.

The devices described above may be implemented as hardware components, software components, and/or a combination of hardware components and software components. For example, devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), It may be implemented using one or more general purpose or special purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. A processing device may run an operating system (OS) and one or more software applications running on the operating system. A processing device may also access, store, manipulate, process, and generate data in response to execution of software. For convenience of understanding, there are cases in which one processing device is used, but those skilled in the art will understand that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it can include. For example, a processing device may include a plurality of processors or a processor and a controller. Other processing configurations are also possible, such as parallel processors.

Software may include a computer program, code, instructions, or a combination of one or more of the foregoing, which configures a processing device to operate as desired or processes independently or collectively. You can command the device. Software and/or data may be any tangible machine, component, physical device, virtual equipment, computer storage medium or device, intended to be interpreted by or provide instructions or data to a processing device. , or may be permanently or temporarily embodied in a transmitted signal wave. Software may be distributed on networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer readable media.

10 is a flowchart illustrating a method for recognizing a work motion based on image analysis according to an embodiment of the present invention.

The video analysis-based work motion recognition method described here is only one embodiment of the present invention, and other than that, various steps may be added as follows, and the following steps may be performed by changing the order. , The present invention is not limited to the steps and their order described below.

Referring to FIGS. 1 and 10 , in step 1010, the work image collection unit 110 of the image analysis-based work motion recognition device 100 may acquire image data about a work site captured through a camera. .

Next, in step 1020, the work motion recognition unit 120 of the image analysis-based work motion recognition device 100 may extract feature information for each frame from the image data.

Next, in step 1030, the work motion recognition unit 120 of the image analysis-based work motion recognition device 100 analyzes the extracted feature information for each frame to classify the work actions constituting one process into semantic units. can be distinguished.

Next, in step 1040, the work motion recognition unit 120 of the video analysis-based work motion recognition device 100 subdivides the work motions of workers in the work site based on the work motions classified by the semantic units. can be recognized.

Next, in step 1050, the work motion recognition unit 120 of the image analysis-based work motion recognition device 100 may derive a sequence of the subdivided work motions.

Next, in step 1060, the task management unit 130 of the image analysis-based work motion recognition apparatus 100 may perform labeling for each video section based on the order of the subdivided work motions.

Next, in step 1070, the task management unit 130 of the video analysis-based task motion recognition device 100 classifies the subdivided task motions by the video section through the labeling and outputs them to the screen of the monitoring terminal. have.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program commands recorded on the medium may be specially designed and configured for the embodiment or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CDROMs and DVDs, and magnetic-optical media such as floptical disks. Included are hardware devices specially configured to store and execute program instructions, such as magneto-optical media and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, as well as machine language codes such as those produced by a compiler. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

As described above, although the embodiments have been described with limited examples and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques may be performed in an order different from the method described, and/or components of the described system, structure, device, circuit, etc. may be combined or combined in a different form than the method described, or other components may be used. Or even if it is replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents of the claims are within the scope of the following claims.

100: image analysis-based work motion recognition device
110: work image collection unit
120: work motion recognition unit
130: work management unit
140: control unit
210: feature information extraction module
220: motion sequence analysis module

Claims (12)

Work image collection unit for obtaining image data on the work site taken by the camera; and
A work motion recognition unit extracting feature information for each frame from the image data and analyzing the extracted feature information for each frame to recognize a work motion of a worker in the work site;
The work action recognition unit analyzes the feature information for each frame to classify work actions constituting one process into semantic units, and subdivides the work actions from the feature information for each frame based on the work actions classified into each semantic unit. and recognizes it, and includes an action sequence analysis module for deriving the order of the subdivided work actions,
Further comprising a task management unit that performs labeling for each video section based on the sequence of the subdivided work operations, classifies the subdivided work operations for each video section through the labeling, and outputs them to a screen of a monitoring terminal,
The work management department
The subdivided working motions of the workers can be realized through job monitoring to identify unnecessary motions of the workers, or to recognize and display which jobs are performed for each section of the video to be used in content production for education and training. It is arranged according to the work order and output on the screen of the monitoring terminal as an analysis result, but whenever work is completed, task management data including the current time, instrument-specific ID value, operator-specific ID value, and work required time are transmitted to the monitoring terminal Image analysis-based work motion recognition device, characterized in that.
According to claim 1,
The work motion recognition unit
Extracting feature information including at least one of joint angle, joint position, torso position, hand position, and line of sight from the image data using at least one technique of joint recognition, hand position recognition, and semantic segmentation Feature Information Extraction Module
Image analysis-based work motion recognition device comprising a.
According to claim 2,
The feature information extraction module
When the worker works using the equipment, the video analysis-based work motion recognition device, characterized in that for extracting the feature information further using a technique for recognizing at least one of the movement and sound of the equipment.
delete According to claim 1,
The operation sequence analysis module
An image analysis-based work motion recognition device comprising a classifier for determining the subdivided work motions and sequentially stacking each subdivided work motion to generate a work motion sequence.
According to claim 5,
The classifier
It is derived by learning the deep learning model through the data of the learning data library,
The learning data library
An image analysis-based work motion recognition device, characterized in that a library that matches detailed motion names, detailed motion work images, and feature information for each frame and stores them as data related to the subdivided work motions.
According to claim 6,
The classifier
The input feature information for each frame is processed to generate an output detailed action name, and a detailed action name matching the input feature information for each frame is extracted from the learning data library and compared with the output detailed action name. An image analysis-based task motion recognition device, characterized in that performing a weight update when a loss occurs according to the result.
According to claim 5,
The classifier
An image analysis-based work motion recognition device, characterized in that the relationship between the feature information for each frame and the subdivided work motion is derived by identifying the rule-based.
delete In the image analysis-based work motion recognition method using an image analysis-based work motion recognition device,
Acquiring image data related to a work site captured by a working image collection unit of the image analysis-based work motion recognition device through a camera;
extracting feature information for each frame from the image data by a work motion recognition unit of the image analysis-based work motion recognition device; and
Recognizing the work motion of the operator in the work site by the work motion recognition unit analyzing the extracted feature information for each frame;
Recognizing the work motion may include analyzing the feature information for each frame and classifying the work motion constituting one process into semantic units; segmenting and recognizing the work motion from the feature information for each frame based on the work motions classified for each semantic unit; And deriving a sequence of the subdivided work operations,
The task management unit of the video analysis-based work motion recognition device performs labeling for each video section based on the sequence of the subdivided work motions, and classifies the subdivided work motions for each video section through the labeling to screen the monitoring terminal. Further comprising the step of outputting to
The work management department
The subdivided working motions of the workers can be realized through job monitoring to identify unnecessary motions of the workers, or to recognize and display which jobs are performed for each section of the video to be used in content production for education and training. It is arranged according to the work order and output on the screen of the monitoring terminal as an analysis result, but whenever work is completed, task management data including the current time, instrument-specific ID value, operator-specific ID value, and work required time are transmitted to the monitoring terminal Image analysis-based work motion recognition method, characterized in that.
According to claim 10,
The step of extracting feature information for each frame from the image data
Extracting feature information including at least one of joint angle, joint position, torso position, hand position, and line of sight from the image data using at least one technique of joint recognition, hand position recognition, and semantic segmentation step
Image analysis-based work motion recognition method comprising a.
delete

Images