CN-121999536-A - Intelligent monitoring system and method for abnormal behavior identification

CN121999536ACN 121999536 ACN121999536 ACN 121999536ACN-121999536-A

Abstract

The invention relates to the field of behavior monitoring and discloses an intelligent monitoring system and method for abnormal behavior identification, wherein the intelligent monitoring method for abnormal behavior identification comprises the steps of continuously collecting the position relations of limbs, an operation tool and a workpiece of personnel, executing asynchronous action semantic splitting on video data, introducing a variable time stepping rule matched with a station rhythm, recombining and correlating cross-frame visual information, extracting a continuous operation characterization unit reflecting an operation propulsion state, carrying out joint evaluation on the offset degree between a current operation process and a standard operation track on the basis of the continuous operation characterization unit, mapping abnormal behavior fragments back to a corresponding station space range, carrying out local reduction on the interaction state among the personnel, the tool and the workpiece, and carrying out dynamic correction on the sampling frequency, the monitoring attention weight and an abnormal judgment boundary of a key vision. The invention has the advantage of improving the accuracy of identifying abnormal behaviors in station operation.

Inventors

BAI YANG
LU JINXIN
CAO LIJUN

Assignees

宁波互诚企业服务有限公司

Dates

Publication Date: 20260508
Application Date: 20260331

Claims (10)

1. An intelligent monitoring method for abnormal behavior identification is characterized by comprising the following steps: Continuously collecting the position relations of the limbs, the operating appliances and the workpieces, and executing asynchronous action semantic splitting on the video data according to the process beats and the operation stage boundaries to form staggered operation observation data; Aiming at operation observation data, introducing a variable time stepping rule matched with the station rhythm, recombining and correlating the cross-frame visual information, and extracting a continuous operation characterization unit reflecting the operation propulsion state by combining the hand movement smoothness, the tool and workpiece contact continuous characteristics and the visual field shielding change trend; on the basis of a continuous operation characterization unit, carrying out joint evaluation on the deviation degree between the current operation process and the standard operation track, and positioning an abnormal behavior segment with operation range-losing risk by capturing structural abnormal changes; Mapping the abnormal behavior segments back to the corresponding station space range, locally reducing the interaction state among the related personnel, appliances and workpieces, and constructing a station interaction disturbance relation structure for describing the propagation path influenced by the abnormal behavior by combining the station layout and the safety constraint relation; Based on the station interactive disturbance relation structure, combining the real-time operation intensity and the past abnormal occurrence distribution, dynamically correcting the sampling frequency, the monitoring attention weight and the abnormal judgment boundary of the key vision.
2. The intelligent monitoring method for abnormal behavior recognition according to claim 1, wherein the process of forming the operation observation data in a staggered arrangement is as follows: acquiring continuous operation pictures through a multi-channel video acquisition device arranged at a station key visual angle, and synchronously positioning key points of limbs, appliance outlines and workpiece boundaries of personnel; Based on predefined operation beat parameters and operation phase division rules in the technical regulations, carrying out phase alignment analysis on the continuous video stream; introducing an asynchronous segmentation strategy at a phase boundary, and splitting and recombining the inter-phase action fragments; And organizing the split action fragments according to the time index and the space identifier, and outputting operation observation data which are arranged in a staggered manner.
3. The intelligent monitoring method for abnormal behavior recognition according to claim 2, wherein the process of reorganizing and correlating the cross-frame visual information is as follows: according to the working position operation rhythm parameters, corresponding time step scales are distributed to each action segment in the operation observation data; Adopting short-time stepping for the high-frequency fragments and long-time stepping for the low-frequency fragments to form a multi-scale time window set; Performing time stamp alignment and sequence correction on the cross-frame visual information in each time window, and eliminating the influence of acquisition delay and inter-frame dislocation; And establishing an action segment association relationship between adjacent time windows to finish the recombination association of the cross-frame visual information.
4. The intelligent monitoring method for identifying abnormal behavior according to claim 3, wherein the process of extracting the continuous operation characterization unit reflecting the operation propulsion state comprises the following steps: Extracting hand motion tracks of the personnel from the recombined cross-frame visual information, and calculating the speed continuity and the direction change rate of the tracks; continuously detecting a contact area between the tool and the workpiece, and counting the contact start-stop time and the stability duration to form a contact continuous characteristic; Comparing the shielding relation among the personnel, the appliances and the workpieces in the operation view field, and extracting the shielding range and the variation trend; And combining the hand movement characteristic, the contact duration characteristic and the shielding change characteristic to construct a continuous operation characterization unit for describing the operation pushing state.
5. The intelligent monitoring method for identifying abnormal behavior according to claim 4, wherein the process of jointly evaluating the deviation degree between the current operation process and the standard operation track is as follows: Standard operation track data corresponding to the current station and the current working procedure are called from a standard operation library, and alignment processing is carried out on the continuous operation characterization unit and the standard operation track in the time dimension and the space dimension; And respectively calculating an action sequence offset, an attitude change offset and an operation rhythm offset, and carrying out weighted fusion on various offsets to obtain a comprehensive offset evaluation result reflecting the current operation deviation degree.
6. The intelligent monitoring method for identifying abnormal behavior according to claim 5, wherein the process of locating the abnormal behavior segment with the risk of job miss-range is as follows: Continuously analyzing the comprehensive offset evaluation result along a time axis, and identifying a time section in which the offset is suddenly increased or continuously abnormal; Detecting action rhythm disorder, gesture switching abnormality and operation path divergence change characteristics in a time section, and marking a corresponding operation fragment as a potential abnormal behavior fragment when the structural change characteristics meet preset abnormality judgment conditions; and merging the adjacent potential abnormal behavior fragments, and outputting the abnormal behavior fragments with the risk of job disrange.
7. The intelligent monitoring method for abnormal behavior recognition according to claim 6, wherein the process of locally restoring the interaction state among the related personnel, appliances and workpieces is as follows: determining a corresponding station space range when the abnormality occurs according to the time mark and the visual angle information of the abnormal behavior segment; Extracting the limb key parts of operators, the instruments involved in operation and the spatial distribution state of the processed workpiece in the abnormal occurrence period; reconstructing the contact relation, the relative position relation and the action sequence among three operation elements of personnel, appliances and workpieces, and outputting a local interactive reduction result corresponding to the abnormal behavior segment.
8. The intelligent monitoring method for abnormal behavior recognition according to claim 7, wherein the process of constructing a station interaction disturbance relation structure for describing the propagation path of the abnormal behavior is as follows: Acquiring space layout information of a station and a corresponding safety operation constraint rule, mapping a local interactive reduction result into the station layout, and identifying key personnel, appliances and workpiece nodes related to abnormal behaviors; establishing influence association between nodes according to the operation dependency relationship and the safety constraint relationship between the nodes; and organizing the influence association, and constructing a station interaction disturbance relation structure for describing the abnormal behavior propagation path.
9. The intelligent monitoring method for abnormal behavior recognition according to claim 8, wherein the process of dynamically correcting the sampling frequency, the monitoring attention weight and the abnormal decision boundary of the key vision is as follows: Identifying key nodes and paths with great influence on abnormal diffusion according to the station interaction disturbance relation structure; Combining the current station operation intensity parameter and the historical abnormal occurrence frequency, and distributing risk weights to the key nodes; and dynamically adjusting the video sampling frequency of the corresponding view field and the monitoring attention weight according to the risk weight, and synchronously correcting the abnormal judgment boundary parameters.
10. An intelligent monitoring system for identifying abnormal behavior, which is applied to the intelligent monitoring method for identifying abnormal behavior according to any one of claims 1 to 9, and is characterized by comprising: The action splitting module is used for executing asynchronous action semantic splitting on the video data according to the process beat and the operation stage boundary; The reorganization association module reorganizes and associates the cross-frame visual information and extracts a continuous job characterization unit reflecting the job propulsion state; the offset evaluation module is used for carrying out joint evaluation on the offset degree between the current operation process and the standard operation track and positioning an abnormal behavior segment with operation error range risk; the interaction reduction module is used for mapping the abnormal behavior segments back to the corresponding station space range and carrying out local reduction on the interaction state among the related personnel, appliances and workpieces; and the dynamic correction module is used for dynamically correcting the key vision field by combining the real-time operation intensity and the past abnormal occurrence distribution.

Description

Intelligent monitoring system and method for abnormal behavior identification Technical Field The invention relates to the field of behavior monitoring, in particular to an intelligent monitoring system and method for abnormal behavior identification. Background On an intelligent manufacturing production line, the video monitoring system is widely applied to links such as equipment operation monitoring, operation flow standard constraint, personnel safety management and the like, and automatic recognition and early warning of abnormal behaviors are realized by analyzing image information of personnel and equipment states in an operation area. However, in a production station scene with a fixed beat and a highly standardized operation flow, the existing abnormal behavior identification technology based on video images still has the problems of insufficient identification precision and practicality. In typical stations such as assembly, spot inspection or feeding, the action amplitude of operators is usually small, and partial nonstandard operation behaviors often occur in short-time, partial posture shift or hand track change forms, such as taking and placing workpieces in an out-of-order manner, performing auxiliary operations when equipment is not completely stopped, and the like. The behaviors do not deviate from the normal operation rhythm obviously in the early stage of occurrence, the overall motion characteristics of the behaviors in the video image are highly similar to the compliance operation, the prior art is mostly dependent on the action amplitude threshold value or the coarse granularity behavior template for recognition, the fine anomalies are difficult to distinguish effectively, and the condition of missed detection is easy to occur. Therefore, it is necessary to design an intelligent monitoring system and method for identifying abnormal behaviors, which improves the accuracy of identifying abnormal behaviors in the station operation. Disclosure of Invention Aiming at the defects of the prior art, the invention provides an intelligent monitoring system and method for identifying abnormal behaviors, which have the advantages of improving the accuracy of identifying the abnormal behaviors in station operation and solve the problems in the background art. In order to achieve the purpose of improving the accuracy of identifying abnormal behaviors in station operation, the invention provides the following technical scheme that the intelligent monitoring method for identifying abnormal behaviors comprises the following steps: Continuously collecting the position relations of the limbs, the operating appliances and the workpieces, and executing asynchronous action semantic splitting on the video data according to the process beats and the operation stage boundaries to form staggered operation observation data; Aiming at operation observation data, introducing a variable time stepping rule matched with the station rhythm, recombining and correlating the cross-frame visual information, and extracting a continuous operation characterization unit reflecting the operation propulsion state by combining the hand movement smoothness, the tool and workpiece contact continuous characteristics and the visual field shielding change trend; on the basis of a continuous operation characterization unit, carrying out joint evaluation on the deviation degree between the current operation process and the standard operation track, and positioning an abnormal behavior segment with operation range-losing risk by capturing structural abnormal changes; Mapping the abnormal behavior segments back to the corresponding station space range, locally reducing the interaction state among the related personnel, appliances and workpieces, and constructing a station interaction disturbance relation structure for describing the propagation path influenced by the abnormal behavior by combining the station layout and the safety constraint relation; Based on the station interactive disturbance relation structure, combining the real-time operation intensity and the past abnormal occurrence distribution, dynamically correcting the sampling frequency, the monitoring attention weight and the abnormal judgment boundary of the key vision. Preferably, the process of forming the staggered operation observation data is as follows: acquiring continuous operation pictures through a multi-channel video acquisition device arranged at a station key visual angle, and synchronously positioning key points of limbs, appliance outlines and workpiece boundaries of personnel; Based on predefined operation beat parameters and operation phase division rules in the technical regulations, carrying out phase alignment analysis on the continuous video stream; introducing an asynchronous segmentation strategy at a phase boundary, and splitting and recombining the inter-phase action fragments; And organizing the split action fragments according to the time index a