A method for accurate recognition of human actions applicable to VR training in grass-roots depots and stations

Abstract

The present invention discloses a method for accurately identifying human actions suitable for VR training at grassroots warehouses and stations, relates to the technical field of detection technology at grassroots warehouses and stations, and solves the technical problem in the prior art that it is impossible to identify and detect the motion trajectories of trainees by differentiating scenes. Specifically, after the trainee's actions are accurately identified, a safety scene is used as the operating environment of the modeling scene to perform action recognition evaluation on the trainee, and the feasibility of the trainee's action execution is inferred through the operation recognition evaluation of each link under the safety scene; and a risk scene is used as the operating environment of the modeling scene to perform the trainee's response action recognition evaluation when any equipment in the area operates abnormally, and infer whether the trainee's response measures for each link in the grassroots warehouse and station meet actual needs.

Description

Technical Field

[0001] The present invention relates to the field of grassroots warehouse station detection technology, and specifically to a method for accurately recognizing human movements suitable for VR training in grassroots warehouse stations. Background Art

[0002] At grassroots depots in the energy, chemical and other industries, daily operations involve a large number of high-risk operational processes, such as oil loading and unloading, chemical storage and transshipment, etc. Traditional training methods rely heavily on theoretical explanations and on-site demonstrations, making it difficult for trainees to fully absorb complex knowledge within a limited time. Furthermore, practical exercises in real-world scenarios can easily lead to safety accidents if errors are made, resulting in immeasurable losses.

[0003] However, in the existing technology, it is impossible to perform modeling control during training at grassroots warehouses and stations, that is, it is impossible to perform post-modeling operation logic fitting settings, which reduces the efficiency of training. In addition, it is impossible to identify the trainees' misjudgments of actions, resulting in the inability to identify and collect action screening, affecting the accuracy of the training results. At the same time, it is impossible to differentiate the scenes to identify and detect the trainees' action trajectories, and it is impossible to effectively infer the feasibility of action execution at grassroots warehouses and stations.

[0004] In view of the above technical defects, a solution is now proposed. Summary of the Invention

[0005] The purpose of the present invention is to solve the above-mentioned problems and to propose a method for accurately recognizing human actions suitable for VR training in grassroots libraries and stations.

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

[0007] A method for accurately identifying human actions suitable for VR training at grassroots libraries and stations. The specific process of the method is as follows:

[0008] S1. VR modeling of grassroots depots and stations: 3D modeling of the areas within the grassroots depots and stations and the equipment within them. During the modeling phase, the equipment within the areas is scaled proportionally to the actual working scene. After the modeling is completed, detailed modeling of the construction coordination area is carried out, and the coordination operation logic of the equipment in each area is aligned to ensure that personnel undergoing VR training can perform actions in accordance with the actual scene;

[0009] S2, action misjudgment identification, after the modeling is completed, according to the actual modeling scene, the trainees should perform actions according to the scene, and the trainees' action misjudgment identification is performed;

[0010] S3. Safety scenario action recognition evaluation: After the trainee's actions are accurately recognized, the safety scenario is used as the operating environment for the modeling scenario to conduct action recognition evaluation on the trainee. By evaluating the operation recognition of each link in the safety scenario, the feasibility of the trainee's action execution is inferred;

[0011] S4. Risk scenario action recognition assessment: using risk scenarios as the operating environment for modeling scenarios, conduct training personnel’s response action recognition assessment when any equipment in the area operates abnormally, and infer whether the training personnel’s response measures for each link in the grassroots warehouse and station meet the actual needs.

[0012] As a preferred embodiment of the present invention, the VR modeling process of the grassroots warehouse station is as follows:

[0013] Modeling and layout planning are carried out based on the actual layout of each area within the grassroots depot, and the equipment in each area of the grassroots depot is simulated to restore the appearance and internal structure of the equipment;

[0014] The operation logic of each area in the grassroots warehouse station under the actual scenario is obtained, and the operation actions of the adjacent areas are set as operation sub-logic, and the operation sub-logic constructs the entire operation logic according to the operation process of the grassroots warehouse station; the preset operation logic of the equipment is constructed according to the process execution sequence of each area in the grassroots warehouse station, and the operation sub-logic in the actual scenario is compared. If there is a logical deviation between the actual preset operation logic and the actual operation sub-logic, the logic of the mutual coordination of the equipment in the actual operation sub-logic is analyzed. If there is no deviation in the mutual coordination, the current actual operation sub-logic is marked as production sub-logic, and the actual preset operation logic is marked as ideal logic.

[0015] As a preferred embodiment of the present invention, the execution time is set to different cycle progress in the VR modeling scenario, and the actual operation logic of the trainees is monitored. If the current cycle progress is in the stage of rushing to meet the deadline, that is, the remaining progress time is lower than the preset time of the remaining workload, the trainees actively change the operation logic, that is, it is inferred that the trainees' logical operation is qualified. On the contrary, if the trainees still use the ideal logic, it is inferred that the timing personnel logic transformation is inefficient; if the current cycle progress is in the stage of early completion, the trainees only execute the ideal logic, that is, it is inferred that the trainees' logical operation is theoretical; the trainees' training is summarized according to the logic execution type, and displayed through the human-computer interaction terminal after the training is completed.

[0016] As a preferred embodiment of the present invention, the action misjudgment identification process is as follows:

[0017] In a simulation modeling scenario, the trainee's execution actions are detected. The trainee's limb joints are monitored for displacement using sensors, and the trainee's body movements are marked as execution actions. First, the corresponding operation logic is derived based on the trainee's location, and the operation sub-logic is obtained based on the device that the trainee needs to operate at the current moment.

[0018] Analyze whether the current operation sub-logic receives the execution instruction;

[0019] If yes, the trainee is currently in the instruction phase. The trainee's action execution time and instruction reception time are collected, and the execution delay is obtained based on the time comparison. If the current execution delay is within the delay range and the limb movement has no reset trajectory, the corresponding execution action is marked as a logical action; conversely, if the current execution delay is not within the delay range or the limb movement has a reset trajectory, the corresponding execution action is marked as an illogical action.

[0020] If no, the trainee is currently in the no-command stage. The trainee's real-time body position is collected. If the real-time body position moves, and the moving direction is not in the direction of the location of the operating equipment of the process to be executed in the area, the trainee's current body movement is considered a superfluous instinctive movement; if the real-time body position moves in the direction of the equipment in the area, and the body position enters the area where the equipment is located, the trainee's current body movement is considered an execution deviation action;

[0021] Redundant instinctive actions and illogical actions are uniformly marked as recorded unrecognized actions; execution deviation actions and logical actions are uniformly marked as recorded recognized actions, and the action types of trainees are identified and collected.

[0022] As a preferred embodiment of the present invention, the safety scene action recognition and evaluation process is as follows:

[0023] Set the simulation modeling scenario for trainees to be a safe scenario, that is, the equipment in each area of the simulation modeling scenario cooperates with each other properly or the operating parameters of the equipment in the area are within the rated range of the equipment;

[0024] Collect execution buffer data and buffer impact data;

[0025] If the execution buffer data exceeds the action buffer time threshold, or the buffer impact data exceeds the speed reciprocating floating span threshold, the correct execution process of the current operation logic is compared with the parameters of the real-time execution action, and the abnormalities are marked. After the training of the trainees is completed, the training is displayed through the human-computer interaction terminal;

[0026] If the execution buffer data does not exceed the action buffer time threshold, and the buffer impact data does not exceed the speed reciprocating floating span threshold, the correct execution action process of the current operation logic and the parameters of the real-time execution action are compared and collected, and displayed through the human-computer interaction terminal after the trainees complete the training.

[0027] As a preferred embodiment of the present invention, the execution buffer data is the action buffer duration of the adjacent record recognition action execution moments of the trainee in the safety scenario; the buffer impact data is the reciprocating floating span of the execution speed of the current record recognition action after the trainee generates the action buffer duration during the training stage.

[0028] After the training, the trainees will correct the execution actions in the safety scenario. For example, if the execution actions are marked as abnormal, they will be rectified. If the execution actions are normal but the execution parameters are lower than the parameters of the correct execution actions, they will summarize the execution actions and adjust the execution actions.

[0029] As a preferred embodiment of the present invention, the risk scenario action recognition and assessment process is as follows:

[0030] The simulation modeling scenario for trainees is set as a risk scenario, that is, the equipment in each area of the simulation modeling scenario fails to cooperate with each other or the operating parameters of the equipment in the area are not within the rated range of the equipment. The risk scenario is divided into a safe transition risk stage and a post-transition risk continuation stage;

[0031] Collect risk impact information and risk persistence information;

[0032] If the risk impact information exceeds the change threshold, or the risk persistence information exceeds the decline rate threshold, the currently executed sub-logic will be marked as abnormal and displayed through the human-computer interaction terminal after the trainees complete the training;

[0033] If the risk impact information does not exceed the change threshold, and the risk persistence information does not exceed the decline rate threshold, it is judged that the trainee's execution action in the risk scenario is normal, and the current execution sub-logic is marked as normal. After the trainee's training is completed, it is displayed through the human-computer interaction terminal after the training is completed.

[0034] As a preferred embodiment of the present invention, the risk impact information is the change in the action trajectory corresponding to the operation sub-logic currently executed by the trainee during the safety transition risk stage;

[0035] The risk persistence information refers to the execution logic of the trainees during the risk persistence stage after the transformation, involving the rate of decline of the qualified rate of equipment in the area.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. In this invention, 3D modeling is performed on the areas within the grassroots warehouse and the equipment within the areas. During the modeling phase, the equipment within the areas is scaled proportionally with the actual working scenes. This improves the detection accuracy of VR training through restoration. At the same time, after the modeling is completed, the construction coordination area is finely modeled, and the coordination operation logic of the equipment in each area is aligned. This ensures that personnel undergoing VR training can perform actions in accordance with the actual scenes, effectively improving the training efficiency.

[0038] After completing the modeling, according to the actual modeling scenario, the trainees should take actions according to the scenario, and the trainees' actions will be identified and recognized, that is, the trainees' execution actions will be detected to avoid invalid actions being identified during the training stage, thereby reducing the proportion of parameter trainees' action recognition detection, causing trainees to perform actions unrelated to the grassroots warehouse and station operations, affecting the training results.

[0039] 2. In the present invention, after the trainee's action is accurately identified, the safety scene is used as the operating environment of the modeling scene to conduct action recognition and evaluation of the trainee. By evaluating the operation recognition of each link in the safety scene, the feasibility of the trainee's action execution is inferred, so as to infer the qualified efficiency of the trainee's action execution in the safety scene, ensure the trainee's operational maturity in each area and link, and enable timely action evaluation and rectification of abnormal actions in each link through training;

[0040] Taking the risk scenario as the operating environment of the modeling scenario, the trainees' response actions are identified and evaluated when any equipment in the area operates abnormally, and it is inferred whether the trainees' response measures for each link in the grassroots warehouse and station meet the actual needs, so as to avoid the trainees' inability to respond in a timely and accurate manner when the operating environment is abnormal. Through risk scenario simulation, the trainees' actions can be effectively detected, and the handling methods under the current risk scenario can be monitored and optimized in real time to reduce the impact of the actual operation risks of the grassroots warehouse and station. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] To facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0042] Figure 1 A flow chart of the overall method of the present invention;

[0043] Figure 2 This is a flow chart of the method for VR modeling of grassroots libraries in the present invention;

[0044] Figure 3 This is a flow chart of the method for identifying action errors in the present invention. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0046] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute a separate or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0047] See also Figure 1 As shown in the figure, a method for accurately recognizing human actions suitable for VR training at grassroots libraries and stations is presented. The specific process of the method is as follows:

[0048] S1. VR modeling of grassroots depots and stations: 3D modeling of the areas within the grassroots depots and stations and the equipment within them. During the modeling phase, the equipment within the area is scaled proportionally with the actual working scene. This improves the detection accuracy of VR training through restoration. At the same time, after the modeling is completed, the construction coordination area is finely modeled, and the coordination operation logic of the equipment in each area is aligned to ensure that personnel undergoing VR training can perform actions in accordance with the actual scene, effectively improving the training efficiency.

[0049] S2, action misjudgment identification. After the modeling is completed, according to the actual modeling scenario, the trainees should perform actions according to the scenario, and the trainees are subjected to action misjudgment identification. That is, the trainees are tested for their actions to avoid invalid actions being identified during the training phase. This reduces the proportion of parameter trainees' action recognition detection, which may cause the trainees to perform actions unrelated to the grassroots warehouse station operations, affecting the training results.

[0050] S3. Safety scenario action recognition assessment: After the trainee's actions are accurately identified, the safety scenario is used as the operating environment for the modeling scenario to conduct action recognition assessment on the trainee. By evaluating the operation recognition of each link in the safety scenario, the feasibility of the trainee's action execution is inferred, so as to infer the qualified efficiency of the trainee's action execution in the safety scenario. This ensures the trainee's operational maturity in each area and link, and enables timely action assessment and rectification of abnormal actions in each link through training;

[0051] S4. Risk scenario action recognition assessment: using risk scenarios as the operating environment for modeling scenarios, conduct training personnel response action recognition assessment when any equipment in the area operates abnormally, infer whether the training personnel's response measures for each link in the grassroots warehouse station meet the actual needs, and avoid the training personnel's inability to respond in a timely and accurate manner due to abnormal operating environment. Through risk scenario simulation, the training personnel's actions can be effectively detected, and the processing methods under the current risk scenario can be monitored and optimized in real time to reduce the impact of the actual operation risks of the grassroots warehouse station.

[0052] See also Figure 2 As shown in the figure, the VR modeling process of S1 grassroots depot station is as follows:

[0053] Modeling and layout planning are carried out based on the actual layout of each area in the grassroots depot, and the equipment in each area of the grassroots depot is simulated to restore the appearance and internal structure of the equipment. Areas such as the oil tank area, loading and unloading platform, pump room, and distribution room in the actual grassroots depot are also planned; equipment such as oil storage tanks, oil pumps, valve groups, etc. are also simulated.

[0054] The operation logic of each area in the grassroots warehouse station under the actual scenario is obtained, and the operation actions of the adjacent areas are set as operation sub-logics, and the operation sub-logics construct the entire operation logic according to the operation process of the grassroots warehouse station; the preset operation logic of the equipment is constructed according to the process execution order of each area in the grassroots warehouse station, and the operation sub-logics in the actual scenario are compared. If there is a logical deviation between the actual preset operation logic and the actual operation sub-logic, the logic of the mutual coordination of the equipment in the actual operation sub-logic is analyzed. If there is no deviation in the mutual coordination, the current actual operation sub-logic is marked as production sub-logic, and the actual preset operation logic is marked as ideal logic. It should be explained that the mutual coordination logic of each equipment in the grassroots warehouse station will be subject to the limitations of the actual site or cost during actual execution, and logical scheduling will occur. For example, the loading and unloading platform may load and unload raw materials of multiple execution processes at the same time to reduce the time cost or input cost of loading and unloading; the operation logic is represented by the corresponding coordinated operation of each equipment used in coordination with each other in the area;

[0055] In the VR modeling scenario, the execution time is set to different cycle progress, and the actual operation logic of the trainees is monitored. If the current cycle progress is in the stage of rushing to meet the deadline, that is, the remaining progress time is lower than the preset time of the remaining workload, the trainees will actively change the operation logic, that is, it is inferred that the trainees' logical operation is qualified. On the contrary, if the trainees still use the ideal logic, it is inferred that the timing personnel's logic transformation is inefficient; if the current cycle progress is in the stage of early completion, the trainees will only execute the ideal logic, that is, it is inferred that the trainees' logical operation is theoretical; the trainees' training is summarized according to the logic execution type, and displayed through the human-computer interaction terminal after the training.

[0056] See also Figure 3As shown in Figure 2, the S2 action misjudgment identification process is as follows:

[0057] In a simulation modeling scenario, the trainee's execution action is detected, the displacement of the trainee's limb joints is monitored according to the sensor, and the trainee's limb movements are marked as execution actions. First, the corresponding operation logic is obtained according to the trainee's area, and the operation sub-logic is obtained according to the equipment that the trainee needs to run at the current moment; it is analyzed whether the current operation sub-logic receives the execution instruction. If so, the current trainee is in the instruction stage, the execution time of the trainee's execution action and the instruction reception time are collected, and the execution delay is obtained based on the time comparison. If the current execution delay is within the delay range, and the limb movement execution has no reset trajectory, that is, the limb movement execution generates a trajectory and then returns to the position before the movement, it is called a reset trajectory, and the corresponding execution action is marked as a logical action; on the contrary, if the current execution delay is not within the delay range, or the limb movement execution has a reset trajectory, the corresponding execution action is marked as an illogical action;

[0058] If no, the trainee is currently in the no-command stage. The trainee's real-time body position is collected. If the real-time body position moves, and the direction of movement is not in the direction of the location of the operating equipment of the process to be executed in the area, the trainee's current body movement is considered a redundant instinctive movement; for example, the trainee's active area inspection and other actions;

[0059] If the real-time limb position moves in the direction of the device in the area, and the limb position enters the area where the device is located, the trainee's current limb movement will be regarded as a deviation action;

[0060] Redundant instinctive actions and illogical actions are uniformly marked as recorded unrecognized actions; execution deviation actions and logical actions are uniformly marked as recorded recognized actions, and the action types of trainees are identified and collected.

[0061] The S3 safety scenario action recognition evaluation process is as follows:

[0062] The trainee's simulation modeling scenario is set as a safe scenario, that is, the equipment in each area of the simulation modeling scenario cooperates with each other satisfactorily or the equipment operating parameters in the area are within the rated range of the equipment. The action buffer duration of the trainee's adjacent record recognition action execution moments in the safe scenario is obtained. At the same time, the execution speed reciprocating floating span of the current record recognition action after the trainee generates the action buffer duration during the training phase is collected. The action buffer duration of the trainee's adjacent record recognition action execution moments in the safe scenario and the execution speed reciprocating floating span of the current record recognition action after the trainee generates the action buffer duration during the training phase are marked as execution buffer data and buffer impact data, respectively, and compared with the action buffer duration threshold and the speed reciprocating floating span threshold, respectively:

[0063] If the action buffer duration of the adjacent recorded recognition action execution moments of the trainee in the safety scenario exceeds the action buffer duration threshold, or the reciprocating floating span of the execution speed of the current recorded recognition action after the trainee generates the action buffer duration during the training phase exceeds the reciprocating floating span threshold, it is inferred that the real-time operation logic execution action of the trainee in the current safety scenario is unqualified, and the correct execution process of the current operation logic is compared with the parameters of the real-time execution action, and after anomalies are marked, parameters such as action trajectory, position and speed are collected together, and displayed through the human-computer interaction terminal after the trainee completes the training;

[0064] If the action buffer duration of the trainee's adjacent recorded recognition action execution moments in the safety scenario does not exceed the action buffer duration threshold, and the reciprocating floating span of the execution speed of the current recorded recognition action after the trainee generates the action buffer duration during the training phase does not exceed the reciprocating floating span threshold, it is inferred that the trainee's real-time operation logic execution action in the current safety scenario is qualified, and the correct execution process of the current operation logic is compared and collected with the parameters of the real-time execution action. After the trainee's training is completed, the parameters are displayed through the human-computer interaction terminal after the training is completed;

[0065] After the training, the trainees will correct the execution actions in the safety scenario. For example, if the execution actions are marked as abnormal, they will be rectified. If the execution actions are normal but the execution parameters are lower than the parameters of the correct execution actions, they will summarize the execution actions and adjust the execution actions.

[0066] The S4 risk scenario action recognition and assessment process is as follows:

[0067] The trainee's simulation modeling scenario is set as a risk scenario, that is, the equipment in each area of the simulation modeling scenario fails to cooperate with each other or the operating parameters of the equipment in the area are not within the rated range of the equipment. The risk scenario is divided into a safety transition risk stage and a post-transition risk persistence stage. The change in the action trajectory corresponding to the operation sub-logic currently executed by the trainee in the safety transition risk stage is collected, where the action trajectory is represented by the execution trajectory of the equipment in the area, such as the execution sequence and click frequency of each button on the console corresponding to the trainee's physical movement trajectory, and the change is represented by the change distance of the physical movement trajectory executed by the execution sub-logic;

[0068] The execution logic of the trainees during the risk persistence phase after the acquisition transition involves the rate of decline in the qualified coordination rate of the equipment in the area. It should be explained that qualified coordination means that the equipment operates properly during the coordination process and the efficiency of the equipment's coordination with each other is qualified. Conversely, unqualified coordination means that any of the above parameters is not met.

[0069] The change in the action trajectory corresponding to the sub-logic of the operation currently executed by the trainee during the safety transition risk phase and the rate of decrease in the qualified cooperation rate of the equipment in the area involved in the trainee's execution logic during the risk persistence phase after the transition are marked as risk impact information and risk persistence information, respectively, and compared with the change threshold and the rate of decrease threshold, respectively:

[0070] If the change in the action trajectory corresponding to the sub-logic of the operation currently executed by the trainee during the safety transition risk phase exceeds the change threshold, or if the rate of decline in the qualified coordinated operation of equipment in the area involved in the trainee's execution logic during the risk persistence phase after the transition exceeds the decline rate threshold, the trainee's execution of the action under the risk scenario is judged to be abnormal, the current execution sub-logic is marked as abnormal, and the abnormality is displayed through the human-computer interaction terminal after the trainee completes the training;

[0071] If the change in the action trajectory corresponding to the operation sub-logic currently executed by the trainee in the safety transition risk stage does not exceed the change threshold, and the rate of decrease in the qualified coordinated operation of equipment in the area involved in the trainee's execution logic in the post-transition risk continuation stage does not exceed the rate of decrease threshold, then it is judged that the trainee's execution of the action is normal under the risk scenario, and the current execution sub-logic is marked as normal. After the trainee's training is completed, it will be displayed through the human-computer interaction terminal after the training is completed.

[0072] When the present invention is in use, the grassroots warehouse station VR modeling is carried out, and the area within the grassroots warehouse station and the equipment within the area are 3D modeled. In the modeling stage, the equipment in the area is scaled proportionally with the actual working scene. At the same time, after the modeling is completed, the construction coordination area is finely modeled, and the coordination operation logic of the equipment in each area is fitted to ensure that the personnel conducting VR training can perform actions in accordance with the actual scene; action error judgment recognition, after the modeling is completed, according to the actual modeling scene, the trainees take actions in response to the scene, and the trainees are subjected to action error judgment recognition; safety scene action recognition evaluation, after the trainees' actions are accurately identified, the safety scene is used as the operating environment of the modeling scene, and the trainees' action recognition evaluation is carried out; risk scene action recognition evaluation, using the risk scene as the operating environment of the modeling scene, the trainees' action recognition evaluation is carried out when any equipment in the area operates abnormally.

[0073] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to specific embodiments. Obviously, many modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for accurately recognizing human actions suitable for VR training at grassroots libraries and stations, characterized by: The specific action accurate recognition method process is as follows: S1. VR modeling of grassroots depots and stations: 3D modeling of the areas and equipment within the grassroots depots and stations. During the modeling phase, the equipment within the area is scaled proportionally to the actual working scene. After the modeling is completed, the construction coordination area is finely modeled, and the coordination operation logic of the equipment in each area is aligned to ensure that personnel undergoing VR training can perform actions in accordance with the actual scene. The VR modeling process of grassroots depots and stations is as follows: Modeling and layout planning are carried out based on the actual layout of each area within the grassroots depot, and the equipment in each area within the grassroots depot is simulated to restore the equipment appearance and internal structure; The operation logic of each area in the grassroots depot in the actual scenario is obtained, and the operation actions of the adjacent areas are set as operation sub-logics. The operation sub-logics construct the entire operation logic according to the operation process of the grassroots depot; The preset operation logic of the equipment is constructed based on the process execution sequence of each area in the grassroots warehouse station, and the operation sub-logic in the actual scenario is compared. If there is a logical deviation between the actual preset operation logic and the actual operation sub-logic, the logic of the equipment coordination of the actual operation sub-logic is analyzed. If there is no deviation in the coordination, the current actual operation sub-logic is marked as the production sub-logic, and the actual preset operation logic is marked as the ideal logic. S2, action misjudgment identification, after the modeling is completed, according to the actual modeling scene, the trainees should perform actions according to the scene, and the trainees' action misjudgment identification is performed; S3. Safety scenario action recognition evaluation: After the trainee's actions are accurately recognized, the safety scenario is used as the operating environment for the modeling scenario to conduct action recognition evaluation on the trainee. By evaluating the operation recognition of each link in the safety scenario, the feasibility of the trainee's action execution is inferred; S4. Risk scenario action recognition assessment: using risk scenarios as the operating environment for modeling scenarios, conduct training personnel’s response action recognition assessment when any equipment in the area operates abnormally, and infer whether the training personnel’s response measures for each link in the grassroots warehouse and station meet the actual needs.

2. The method for accurately recognizing human actions suitable for VR training at grassroots libraries and stations according to claim 1 is characterized in that: In the VR modeling scenario, the execution time is set to different cycle progress, and the actual operation logic of the trainees is monitored. If the current cycle progress is in the stage of rushing to meet the deadline, that is, the remaining progress time is lower than the preset time of the remaining workload, the trainees will actively change the operation logic, that is, it is inferred that the trainees' logical operation is qualified. On the contrary, if the trainees still use the ideal logic, it is inferred that the timing personnel's logic transformation is inefficient; if the current cycle progress is in the stage of early completion, the trainees will only execute the ideal logic, that is, it is inferred that the trainees' logical operation is theoretical; the trainees' training is summarized according to the logic execution type, and displayed through the human-computer interaction terminal after the training.

3. The method for accurately recognizing human actions suitable for VR training at grassroots libraries and stations according to claim 1 is characterized in that: The process of identifying misjudgments of actions is as follows: in a simulation modeling scenario, the trainee's execution actions are detected, the displacement of the trainee's limb joints is monitored using sensors, and the trainee's limb movements are marked as execution actions. First, the corresponding operation logic is derived based on the trainee's location, and the operation sub-logic is obtained based on the device that the trainee needs to operate at the current moment. The current operation sub-logic is analyzed to determine whether it has received the execution instruction. If yes, the trainee is currently in the instruction phase. The trainee's action execution time and instruction reception time are collected, and the execution delay is obtained based on the time comparison. If the current execution delay is within the delay range and the limb movement has no reset trajectory, the corresponding execution action is marked as a logical action; conversely, if the current execution delay is not within the delay range or the limb movement has a reset trajectory, the corresponding execution action is marked as an illogical action. If not, the current trainee is in the no-instruction stage, and the trainee's real-time limb position is collected. If the real-time limb position moves, and the moving direction is not in the direction of the location of the operating equipment of the process to be executed in the area, the trainee's current limb movement is considered a redundant instinctive movement; if the moving direction of the real-time limb position is in the direction of the equipment in the area, and the limb position enters the area where the equipment is located, the trainee's current limb movement is considered an execution deviation movement; redundant instinctive movements and illogical movements are uniformly marked as recorded unrecognized movements; execution deviation movements and logical movements are uniformly marked as recorded recognized movements, and the trainee's movement types are identified and collected.

4. The method for accurately recognizing human actions suitable for VR training at grassroots libraries and stations according to claim 1 is characterized in that: The safety scene action recognition evaluation process is as follows: Set the simulation modeling scenario for trainees to be a safe scenario, that is, the equipment in each area of the simulation modeling scenario cooperates with each other properly or the equipment operating parameters in the area are within the rated range of the equipment; collect execution buffer data and buffer impact data; If the execution buffer data exceeds the action buffer time threshold, or the buffer impact data exceeds the speed reciprocating floating span threshold, the correct execution process of the current operation logic is compared with the parameters of the real-time execution action, and the abnormalities are marked. After the training of the trainees is completed, the training is displayed through the human-computer interaction terminal; If the execution buffer data does not exceed the action buffer time threshold, and the buffer impact data does not exceed the speed reciprocating floating span threshold, the correct execution action process of the current operation logic and the parameters of the real-time execution action are compared and collected, and displayed through the human-computer interaction terminal after the trainees complete the training.

5. The method for accurately recognizing human actions suitable for VR training at grassroots libraries and stations according to claim 4 is characterized in that: The execution buffer data is the action buffer duration of the trainee's adjacent record recognition action execution in the safety scenario; the buffer impact data is the reciprocating floating span of the execution speed of the current record recognition action after the trainee generates the action buffer duration; After the training, the trainees will correct the execution actions in the safety scenario. Specifically, they will rectify the execution actions marked as abnormal. If the execution actions are normal but the execution parameters are lower than the parameters of the correct execution actions, they will summarize the execution actions and adjust the execution actions.

6. The method for accurately recognizing human actions suitable for VR training at grassroots libraries and stations according to claim 1 is characterized in that: The risk scenario action recognition and assessment process is as follows: The simulation modeling scenario for trainees is set as a risk scenario. That is, the equipment in each area of the simulation modeling scenario fails to cooperate with each other or the operating parameters of the equipment in the area are not within the rated range of the equipment. The risk scenario is divided into a safe transition risk stage and a post-transition risk continuation stage; risk impact information and risk continuation information are collected; If the risk impact information exceeds the change threshold, or the risk persistence information exceeds the decline rate threshold, the currently executed sub-logic will be marked as abnormal and displayed through the human-computer interaction terminal after the trainees complete the training; If the risk impact information does not exceed the change threshold, and the risk persistence information does not exceed the decline rate threshold, it is judged that the trainee's execution action in the risk scenario is normal, and the current execution sub-logic is marked as normal. After the trainee's training is completed, it is displayed through the human-computer interaction terminal after the training is completed.

7. The method for accurately recognizing human actions suitable for VR training at grassroots libraries and stations according to claim 6 is characterized in that: The risk impact information is the change in the action trajectory corresponding to the operation sub-logic currently executed by the trainee during the safety transition risk stage; The risk persistence information refers to the execution logic of the trainees during the risk persistence stage after the transformation, involving the rate of decline of the qualified rate of equipment in the area.

Images