CN-122024182-A - Transformer substation video monitoring method and system based on three-dimensional model

CN122024182ACN 122024182 ACN122024182 ACN 122024182ACN-122024182-A

Abstract

The invention discloses a three-dimensional model-based substation video monitoring method and system, wherein the method comprises the steps of identifying abnormal behavior track points according to sliding window gradient detection and relative change rate continuous criteria, and determining abnormal behavior data by combining with space distance verification of a safety area of electrified equipment; generating normal and abnormal behavior subsequences according to time continuity, clustering according to length, establishing association relation between the abnormal subsequences and adjacent normal subsequences, selecting target abnormal subsequences, calculating correction parameters, multiplexing the correction parameters in batches to other abnormal subsequences in the same set, completing second-layer correction through space-time behavior field and dynamic safety field coupling, variation mode decomposition periodic interference and behavior topology network confidence weighting, and finally outputting a high-precision correction behavior track. The method solves the problems of inaccurate space positioning and low abnormal recognition reliability of the transformer substation monitoring, and realizes accurate recognition and correction of the movement track of personnel or equipment.

Inventors

WANG YATING
CHENG WENXING
YUAN XINGHUA
JIN MING
CHEN QINGHUA
ZHAN HAIFENG
ZHU YUN
Wang Chugun
LAI DONGSHENG

Assignees

江西腾达电力设计院有限公司
国网江西省电力有限公司经济技术研究院

Dates

Publication Date: 20260512
Application Date: 20260416

Claims (10)

1. The transformer substation video monitoring method based on the three-dimensional model is characterized by comprising the following steps of: acquiring a three-dimensional model of a transformer substation, and establishing a mapping relation between video pixel coordinates of each monitoring camera and space coordinates of the three-dimensional model according to preset camera calibration parameters; Acquiring real-time video streams acquired by at least two monitoring cameras, aligning and fusing three-dimensional space point clouds corresponding to video frames of the at least two monitoring cameras at the same moment according to the mapping relation of each monitoring camera respectively to generate a unified three-dimensional space point cloud sequence; performing anomaly detection on track points in the three-dimensional space point cloud sequence according to a preset behavior anomaly detection strategy, and judging whether abnormal behavior data exist or not; If at least one piece of abnormal behavior data exists, generating at least one normal behavior data subsequence and at least one abnormal behavior data subsequence according to the time continuity of the abnormal behavior data in the three-dimensional space point cloud sequence, and carrying out sequence division on the at least one abnormal behavior data subsequence to obtain at least one abnormal behavior data subsequence set; For each abnormal behavior data subsequence, respectively establishing an association relationship between the abnormal behavior data subsequence and a normal behavior data subsequence adjacent to the abnormal behavior data subsequence on a time node; selecting a target abnormal behavior data subsequence from a certain abnormal behavior data subsequence set, and correcting the target abnormal behavior data subsequence by adopting a preset first behavior correction strategy according to a target association relation containing the target abnormal behavior data subsequence to obtain a certain corrected abnormal behavior data subsequence and a certain behavior correction parameter; directly correcting other abnormal behavior data subsequences in the certain abnormal behavior data subsequence set according to the certain behavior correction parameter to obtain other corrected abnormal behavior data subsequences; And splicing each abnormal behavior data correction sub-sequence with the at least one normal behavior data sub-sequence based on the sequence of the time nodes to obtain a behavior data sequence to be corrected, and correcting the behavior data sequence to be corrected again according to a preset second behavior correction strategy to finally obtain a corrected behavior track of personnel or equipment in the transformer substation monitoring area.
2. The substation video monitoring method based on the three-dimensional model according to claim 1, wherein the generating a unified three-dimensional space point cloud sequence by aligning and fusing three-dimensional space point clouds corresponding to video frames of at least two monitoring cameras at the same moment according to the mapping relation of each monitoring camera comprises: mapping each pixel point in each video frame in each real-time video stream to a three-dimensional model space one by one to obtain a three-dimensional space point cloud corresponding to each video frame; Aligning and fusing three-dimensional space point clouds corresponding to video frames of at least two monitoring cameras at the same moment according to time stamps to generate a unified three-dimensional space point cloud sequence, wherein the method specifically comprises the following steps: Performing time synchronization on the real-time video streams of each monitoring camera to acquire video frames acquired by each monitoring camera at the same moment; According to camera calibration parameters and pose parameters of each monitoring camera, mapping all pixel points in each video frame to a three-dimensional model space one by one to obtain a three-dimensional space point cloud corresponding to each monitoring camera at the same time; Carrying out space registration on three-dimensional space point clouds from different cameras at the same moment, identifying overlapping points by adopting a space index, judging the overlapping points when the space distance between the two points is smaller than a preset distance threshold value, carrying out weighted fusion on the overlapping points, directly merging non-overlapping points, and generating a fused three-dimensional space point cloud at the same moment; and arranging the fusion three-dimensional space point clouds at all the moments in time sequence to generate a unified three-dimensional space point cloud sequence.
3. The three-dimensional model-based substation video monitoring method according to claim 1, wherein the abnormal behavior data is motion trail data of a person or equipment, which deviate from normal behavior and have abnormal spatial topological relation with a safety area of the electrified equipment, in a monitoring area; the step of carrying out anomaly detection on the track points in the three-dimensional space point cloud sequence according to a preset behavior anomaly detection strategy, and the step of judging whether the anomaly behavior data exist comprises the following steps: sequentially acquiring track change gradients between two adjacent points of a target track in the three-dimensional space point cloud sequence based on the sequence of the time nodes to obtain a track change gradient sequence; Sliding on the track change gradient sequence according to a preset sliding window, and judging whether the track change gradient in the sliding window is larger than a preset gradient threshold value or not when sliding each time; if the change gradient of each track is not greater than a preset gradient threshold value, determining that abnormal behavior data does not exist; If a certain track change gradient is larger than a preset gradient threshold value, acquiring the relative change rate between each target track change gradient and the certain track change gradient, wherein the target track change gradient is the track change gradient which is sequenced to be positioned behind the certain track change gradient in the track change gradient sequence; Judging whether the number of the relative change rates continuously larger than a preset change rate threshold is larger than a preset number threshold or not; If the number is greater than a preset number threshold, defining track points corresponding to the certain track change gradient and track points corresponding to the relative change rate continuously greater than the preset change rate threshold as candidate abnormal points; And calculating the space distance between the three-dimensional space coordinates corresponding to the candidate abnormal points and the boundary of the safety area of the live equipment in the three-dimensional model of the transformer substation, and defining the candidate abnormal points as abnormal behavior data when the minimum space distance is smaller than a preset safety distance threshold.
4. The method for monitoring the video of the transformer substation based on the three-dimensional model according to claim 1, wherein the generating at least one normal behavior data subsequence and at least one abnormal behavior data subsequence according to the time continuity of the abnormal behavior data in the three-dimensional space point cloud sequence, and the performing the sequence division on the at least one abnormal behavior data subsequence, to obtain at least one abnormal behavior data subsequence set comprises: traversing each track point in the three-dimensional space point cloud sequence, and marking the states of each track point according to an abnormal detection result, wherein the states comprise a normal state and an abnormal state; Dividing the track points continuously in the abnormal state into an abnormal behavior data subsequence, and dividing the track points continuously in the normal state into a normal behavior data subsequence; Recording the starting time stamp, the ending time stamp and the sequence length of each abnormal behavior data sub-sequence, and recording the starting time stamp, the ending time stamp and the sequence length of each normal behavior data sub-sequence; And acquiring the sequence length of the at least one abnormal behavior data subsequence, and clustering each abnormal behavior data subsequence with the same sequence length to obtain at least one abnormal behavior data subsequence set.
5. The method for monitoring the video of the transformer substation based on the three-dimensional model according to claim 1, wherein for each abnormal behavior data subsequence, respectively establishing an association relationship between the abnormal behavior data subsequence and a normal behavior data subsequence adjacent to the abnormal behavior data subsequence on a time node comprises: acquiring a start time stamp and a termination time stamp of the abnormal behavior data subsequence; searching whether a previous normal behavior data subsequence with a first adjacent termination time stamp and a starting time stamp of the abnormal behavior data subsequence exists in each normal behavior data subsequence, wherein the first adjacent refers to that the difference between the termination time stamp of the previous normal behavior data subsequence and the starting time stamp of the abnormal behavior data subsequence is smaller than a preset time difference threshold; Searching whether a next normal behavior data subsequence with a second adjacent starting time stamp and a second adjacent ending time stamp of the abnormal behavior data subsequence exists in each normal behavior data subsequence, wherein the second adjacent means that the difference between the starting time stamp of the next normal behavior data subsequence and the ending time stamp of the abnormal behavior data subsequence is smaller than a preset time difference threshold; Establishing an association relation according to the search result, which specifically comprises the following steps: When the former normal behavior data subsequence and the latter normal behavior data subsequence are searched at the same time, the abnormal behavior data subsequence, the former normal behavior data subsequence and the latter normal behavior data subsequence are stored in an associated mode, and an association relation comprising three subsequences is formed; When only the former normal behavior data subsequence is found, the abnormal behavior data subsequence and the former normal behavior data subsequence are stored in an associated mode to form an association relation containing the two subsequences, and the latter normal behavior data subsequence is marked as missing; when only the next normal behavior data subsequence is found, carrying out association storage on the abnormal behavior data subsequence and the next normal behavior data subsequence to form an association relation containing the two subsequences, and marking the previous normal behavior data subsequence as missing; and when the abnormal behavior data subsequence is not found, marking the abnormal behavior data subsequence as an isolated abnormality, and not establishing an association relation.
6. The substation video monitoring method based on the three-dimensional model according to claim 1, wherein the selecting a target abnormal behavior data subsequence from a certain abnormal behavior data subsequence set, and correcting the target abnormal behavior data subsequence by adopting a preset first behavior correction policy according to a target association relationship including the target abnormal behavior data subsequence, to obtain a certain corrected abnormal behavior data subsequence, and the certain behavior correction parameter includes: in a certain abnormal behavior data subsequence set, sequencing according to the starting time stamp of each abnormal behavior data subsequence, and selecting the abnormal behavior data subsequence with the starting time stamp positioned at the middle position as a target abnormal behavior data subsequence; acquiring a target normal behavior data subsequence associated with the target abnormal behavior data subsequence according to the target association relationship, and determining the number of the target normal behavior data subsequences; When the target association relation comprises two target normal behavior data subsequences, acquiring the sequence length of the target abnormal behavior data subsequence, and respectively intercepting a first target normal behavior data subsequence segment and a second target normal behavior data subsequence segment which are identical to the sequence length and adjacent to the target abnormal behavior data subsequence in a first target normal behavior data subsequence and a second target normal behavior data subsequence segment, wherein the first target normal behavior data subsequence is a previous normal behavior data subsequence adjacent to the target abnormal behavior data subsequence, and the second target normal behavior data subsequence is a subsequent normal behavior data subsequence adjacent to the target abnormal behavior data subsequence; Calculating a first average position vector of each normal behavior track point in the first target normal behavior data subsequence segment and a second average position vector of each normal behavior track point in the second target normal behavior data subsequence segment, and defining a weighted average value of the first average position vector and the second average position vector as a first dynamic normal behavior benchmark; Generating a dynamic security field according to real-time operation parameters of the electrified equipment in the three-dimensional model of the transformer substation, wherein the dynamic security field takes the electrified equipment as a center, the field intensity is attenuated along with the distance and dynamically adjusted along with the change of the operation state of the equipment, the field intensity function is E (d) =E0.e -αd , E (d) is the field intensity at a space point, d is the distance from the space point to the surface of the electrified equipment, E0 is the reference field intensity of the surface of the equipment, and alpha is an attenuation coefficient; sequentially searching whether a three-dimensional space Euclidean distance between the target abnormal behavior data subsequence and the first dynamic normal behavior reference is smaller than a preset distance threshold value along the forward direction of a time node from front to back, and simultaneously meeting a certain target abnormal behavior track point with a field intensity value larger than the preset field intensity threshold value in a dynamic safety field; when the certain target abnormal behavior track point exists, stopping searching, defining the certain target abnormal behavior track point as a cut-off abnormal behavior track point, and acquiring the cut-off abnormal behavior track point and a first position vector difference value between the target abnormal behavior track points which are directly adjacent to the cut-off abnormal behavior track point along the reverse direction of time; And vector addition is carried out on the first position vector difference value and each reverse target abnormal behavior track point in the target abnormal behavior data subsequence to obtain a certain corrected abnormal behavior data subsequence corresponding to the target abnormal behavior data subsequence, wherein each reverse target abnormal behavior track point is a target abnormal behavior track point of a time node before a time node corresponding to the cut-off abnormal behavior track point.
7. The three-dimensional model-based substation video monitoring method according to claim 6, wherein after determining the number of target normal behavior data subsequences, the method further comprises: when the target association relation only comprises one target normal behavior data subsequence, acquiring the sequence length of the target abnormal behavior data subsequence, and intercepting a target normal behavior data subsequence segment which has the same sequence length and is adjacent to the target abnormal behavior data subsequence from the target normal behavior data subsequence; Calculating a third average position vector of each normal behavior track point in the target normal behavior data subsequence segment, and defining the third average position vector as a second dynamic normal behavior benchmark; and sequentially searching whether a three-dimensional space Euclidean distance between the target abnormal behavior data subsequence and the second dynamic normal behavior reference is smaller than a preset distance threshold value along the forward direction of the time node from front to back, and simultaneously meeting a certain target abnormal behavior track point with a field intensity value larger than the preset field intensity threshold value in a dynamic safety field.
8. The substation video monitoring method based on the three-dimensional model according to claim 1, wherein the certain behavior modification parameter is the first position vector difference value; The step of directly correcting other abnormal behavior data subsequences in the certain abnormal behavior subsequence set according to the certain behavior correction parameter to obtain other corrected abnormal behavior data subsequences comprises the following steps: aligning the other abnormal behavior data subsequences with the target abnormal behavior data subsequences, and determining other abnormal behavior trace points which are cut off and have the same sequencing positions as the abnormal behavior trace points cut off in the other abnormal behavior data subsequences; And vector addition is carried out on the certain behavior correction parameter and each reverse other abnormal behavior track point in the other abnormal behavior data subsequence to obtain other corrected abnormal behavior data subsequences corresponding to the other abnormal behavior data subsequences, wherein each reverse other abnormal behavior track point is other abnormal behavior track points of a time node before the time node corresponding to the cut-off other abnormal behavior track point.
9. The method for video monitoring of a transformer substation based on a three-dimensional model according to claim 6, wherein the re-correcting the behavior data sequence to be corrected according to a preset second behavior correction policy, and finally obtaining a corrected behavior track of a person or a device in the transformer substation monitoring area comprises: Acquiring three-dimensional space coordinates of each track point in the behavior data sequence to be corrected, and constructing a space-time behavior field, wherein the space-time behavior field takes the three-dimensional space coordinates as a field and takes a motion speed vector of a person or equipment at the track point as a field vector; Performing field coupling on the space-time behavior field and the dynamic security field to obtain a coupling field, and calculating comprehensive potential energy at each track point in the coupling field to obtain a comprehensive potential energy sequence, wherein the comprehensive potential energy is a dot product of a space-time behavior field vector and a dynamic constraint field vector; Integrating the comprehensive potential energy sequence along the forward direction from the beginning to the end of the time node to obtain an accumulated potential energy curve, and identifying an abnormal fluctuation segment according to the change rate of the accumulated potential energy curve; For an abnormal fluctuation segment in the accumulated potential energy curve, decomposing the abnormal fluctuation segment into a plurality of eigenmode functions by adopting variation mode decomposition, and identifying periodic components related to the running period of substation equipment in each eigenmode function; Calculating the offset compensation quantity of the track point corresponding to the abnormal fluctuation segment according to the identified periodic component, and superposing the offset compensation quantity to the three-dimensional space coordinate of the corresponding track point to obtain a primary correction behavior track; inputting the primary correction behavior track into a preset behavior topology network, aggregating the characteristics of adjacent nodes through a graph neural network, and outputting track correction confidence, wherein the behavior topology network takes a historical normal behavior track as a node and takes track similarity as an edge; And carrying out weighted optimization on the primary correction behavior track according to the track correction confidence coefficient to obtain the correction behavior track of the personnel or equipment in the transformer substation monitoring area.
10. A three-dimensional model-based substation video monitoring system, comprising: The acquisition module is configured to acquire a three-dimensional model of the transformer substation, and establishes a mapping relation between video pixel coordinates of each monitoring camera and space coordinates of the three-dimensional model according to preset camera calibration parameters; The fusion module is configured to acquire real-time video streams acquired by at least two monitoring cameras, align and fuse three-dimensional space point clouds corresponding to video frames of the at least two monitoring cameras at the same moment according to mapping relations of the monitoring cameras respectively, and generate a unified three-dimensional space point cloud sequence; the judging module is configured to detect abnormality of track points in the three-dimensional space point cloud sequence according to a preset behavior abnormality detection strategy and judge whether abnormal behavior data exist or not; The dividing module is configured to generate at least one normal behavior data subsequence and at least one abnormal behavior data subsequence according to the time continuity of the abnormal behavior data in the three-dimensional space point cloud sequence if at least one abnormal behavior data exists, and to divide the at least one abnormal behavior data subsequence in sequence to obtain at least one abnormal behavior data subsequence set; The association module is configured to respectively establish association relation between the abnormal behavior data subsequence and a normal behavior data subsequence adjacent to the abnormal behavior data subsequence on a time node for each abnormal behavior data subsequence; The first correction module is configured to select a target abnormal behavior data subsequence from a certain abnormal behavior data subsequence set, and correct the target abnormal behavior data subsequence by adopting a preset first behavior correction strategy according to a target association relation containing the target abnormal behavior data subsequence to obtain a certain corrected abnormal behavior data subsequence and a certain behavior correction parameter; The second correction module is configured to directly correct other abnormal behavior data subsequences in the certain abnormal behavior data subsequence set according to the certain behavior correction parameter to obtain other corrected abnormal behavior data subsequences; the output module is configured to splice each abnormal behavior data correction sub-sequence with the at least one normal behavior data sub-sequence based on the sequence of the time nodes to obtain a behavior data sequence to be corrected, and correct the behavior data sequence to be corrected again according to a preset second behavior correction strategy to finally obtain a corrected behavior track of personnel or equipment in the transformer substation monitoring area.

Description

Transformer substation video monitoring method and system based on three-dimensional model Technical Field The invention belongs to the technical field of transformer substation video monitoring, and particularly relates to a transformer substation video monitoring method and system based on a three-dimensional model. Background The transformer substation is used as a key node of the power system, and the safe operation of the transformer substation is directly related to the stability and the power supply reliability of the power grid. Along with the wide application of the video monitoring technology in the transformer substation, how to accurately identify the abnormal behaviors of personnel or equipment from massive video data and discover potential safety hazards in time becomes a problem to be solved urgently. The existing transformer substation video monitoring method mainly relies on two-dimensional image analysis, personnel and equipment are identified through a target detection algorithm, the distance between the personnel and the electrified equipment is calculated, and an alarm is triggered when the distance is smaller than a threshold value. However, such methods suffer from the following drawbacks: Firstly, the two-dimensional image lacks depth information, so that the real space distance between a person and the electrified equipment is difficult to accurately judge, and misjudgment or missed judgment is easy to generate due to perspective projection effect. Secondly, the field of view of the monitoring camera is limited, a single camera is difficult to cover the whole area of the transformer substation, and the space cooperation among multiple cameras is lacked, so that uniform space perception cannot be formed. Thirdly, the existing anomaly detection method is generally based on image analysis at a single moment, so that time sequence continuity of personnel behaviors is ignored, and short disturbance and continuous dangerous behaviors are difficult to distinguish. Fourth, for the abnormal behavior that has already occurred, lack effective correction mechanism, can't carry on the compensation to rebuild to the unusual orbit, lead to the incomplete record of monitoring. Therefore, a transformer substation video monitoring method capable of fusing multi-camera information, performing spatial positioning by using a three-dimensional model, identifying abnormal behaviors based on time sequence analysis and performing track correction is needed to improve the accuracy and the robustness of monitoring. Disclosure of Invention The invention provides a transformer substation video monitoring method and system based on a three-dimensional model, which are used for solving the technical problems of inaccurate spatial positioning, insufficient multi-camera coordination, low abnormal behavior recognition reliability and lack of track correction mechanism in the existing transformer substation video monitoring. In a first aspect, the present invention provides a substation video monitoring method based on a three-dimensional model, including: acquiring a three-dimensional model of a transformer substation, and establishing a mapping relation between video pixel coordinates of each monitoring camera and space coordinates of the three-dimensional model according to preset camera calibration parameters; Acquiring real-time video streams acquired by at least two monitoring cameras, aligning and fusing three-dimensional space point clouds corresponding to video frames of the at least two monitoring cameras at the same moment according to the mapping relation of each monitoring camera respectively to generate a unified three-dimensional space point cloud sequence; performing anomaly detection on track points in the three-dimensional space point cloud sequence according to a preset behavior anomaly detection strategy, and judging whether abnormal behavior data exist or not; If at least one piece of abnormal behavior data exists, generating at least one normal behavior data subsequence and at least one abnormal behavior data subsequence according to the time continuity of the abnormal behavior data in the three-dimensional space point cloud sequence, and carrying out sequence division on the at least one abnormal behavior data subsequence to obtain at least one abnormal behavior data subsequence set; For each abnormal behavior data subsequence, respectively establishing an association relationship between the abnormal behavior data subsequence and a normal behavior data subsequence adjacent to the abnormal behavior data subsequence on a time node; selecting a target abnormal behavior data subsequence from a certain abnormal behavior data subsequence set, and correcting the target abnormal behavior data subsequence by adopting a preset first behavior correction strategy according to a target association relation containing the target abnormal behavior data subsequence to obtain a certain corrected abnormal behav