CN-122023862-A - Power transmission line external damage prevention monitoring system and method based on improved DepthPro depth estimation

CN122023862ACN 122023862 ACN122023862 ACN 122023862ACN-122023862-A

Abstract

The monitoring system comprises a target detection module, a depth estimation module, a minimum spatial distance calculation module, a risk assessment and grading early warning module, and a risk assessment and grading early warning module, wherein the target detection module inputs continuous images I t of a monitoring area, outputs spatial positions and masks of a crane boom, a lifting hook, a pole tower and a power transmission line, the depth estimation module inputs a pixel level depth estimation result image D t , detected boom and power transmission line areas, outputs a minimum spatial distance D min , the minimum spatial distance calculation module inputs an original image frame I t and a detection result B t , outputs a pixel level depth estimation result image D t and a pixel level uncertainty U t , and the risk assessment and grading early warning module inputs a minimum spatial distance D min and a pixel level uncertainty U t and outputs a risk grade and an alarm signal. According to the method, YOLOv a target detection algorithm is adopted, a DepthPro depth estimation model is improved, and a time sequence transducer module is introduced to optimize depth data, so that the problems of false detection, omission detection, unstable depth estimation and the like in the prior art can be effectively solved.

Inventors

FAN ZHONGSHENG
Chen tuo
FAN YUNZHEN
LIN HUI
SUN ZHI
XIE XUEPING
LU QI
CHEN CHENG
YANG FAN
XU LEPING
FENG WEI
CHENG LEI

Assignees

国网湖北省电力有限公司孝感供电公司

Dates

Publication Date: 20260512
Application Date: 20251204

Claims (10)

1. Power transmission line external damage prevention monitoring system based on improvement DepthPro degree of depth estimation, characterized by that this monitoring system includes: The target detection module inputs continuous images I t of the monitoring area and outputs the space positions and masks of a crane boom, a lifting hook, a pole tower and a power transmission line; The depth estimation module is used for inputting a pixel-level depth estimation result graph D t , a detected suspension arm and power transmission line area and outputting a minimum space distance D min ; The minimum space distance calculation module inputs the original image frame I t and the detection result B t , and outputs a pixel-level depth estimation result image D t and a pixel-level uncertainty U t ; And the risk assessment and grading early warning module is used for inputting the minimum spatial distance d min and the pixel level uncertainty U t and outputting risk grades and warning signals.
2. The system for monitoring damage to a power transmission line based on improved DepthPro depth estimation according to claim 1, wherein the object detection module includes: a1, inputting an image sequence { I t } of a monitoring area, wherein the image sequence { I t } comprises a crane boom, a lifting hook, a pole tower and a power transmission line; a2, inputting the image into YOLOv network, and outputting the boundary box and class label ci of the target; b i denotes that the image contains spatial coordinates and category information, i=1, 2,..n, n denotes that the image has n object boxes; a3, for the transmission line part, further outputting a pixel level mask For subsequent depth estimation and spatial relationship modeling.
3. The system for monitoring the damage to the power transmission line based on the improved DepthPro depth estimation according to claim 2, wherein the bounding box and the class label ci are calculated as follows: ; Wherein, B t represents a detection result, namely a target set detected by the moment t, which contains label category, boundary box and confidence si information, and specifically, B t = (ci, si, x_center, y_ cneter, width, height); Wherein, x_center, y_ cetner represents the coordinate of the center point of the target, width represents the transverse distance from the coordinate of the center point to the target frame, height represents the longitudinal distance from the coordinate of the center point to the target frame; representation YOLOv11 network model; b i denotes that the image contains spatial coordinates and class information, i=1, 2,..n, n denotes that there are n object boxes.
4. The power transmission line damage prevention monitoring system based on improved DepthPro depth estimation as set forth in claim 3, wherein the pixel level mask The specific calculation is as follows: ; Wherein, the Representing Sigmoid function, proto representing prototype mask, coef_i representing mask coefficients.
5. The system for monitoring damage to a power transmission line based on improved DepthPro depth estimation as set forth in claim 4, wherein the depth estimation module includes: b1, inputting the image I t into the improved DepthPro model to obtain a pixel level depth estimation result graph D t (ci, si, x_center, y_center, width, height, mt): ; wherein Mt is a pixel level mask, (x_center, y_ cneter, width, height) is a bounding box and ci is a class label; b2, introducing a time sequence transducer module at a prediction end of the DepthPro model, and fusing by utilizing context information of adjacent image frames (I t-1 ,I t ) to relieve single-frame depth drift; b3, outputting pixel-level uncertainty U t for subsequent risk assessment; ; Wherein D t represents a pixel level depth estimation result diagram, U t represents pixel level uncertainty corresponding to depth estimation, TTS represents a time sequence transducer, and Concat is a splicing operation; Representing a depth estimation model.
6. The system for monitoring damage to a power transmission line based on improved DepthPro depth estimation as set forth in claim 5, wherein the minimum spatial distance calculation module includes: c1, combining the detected image coordinates of each pixel in the monitoring area and the corresponding depth value into three-dimensional points according to the detection result B t and the pixel-level depth estimation result graph D t , so as to form a three-dimensional point set representing the part of the boom and the wire, namely a pseudo point cloud (x t ,y t ,z t ) for subsequent spatial analysis: ; Wherein d is the estimated depth value of the pixel point, K is an internal reference matrix, and (u, v) is a two-dimensional pixel coordinate; c2, geometrically correcting the power transmission line region by applying physical prior; ; Wherein z ' represents the estimated height value of the transmission line at the position x ', which is the output of the fitting curve, x ' represents the horizontal distance along the transmission line direction, the projected point of the first tower is usually set as the origin of coordinates; c3, extracting point cloud set data of the suspension arm by using an instrument to obtain P arm ; And c4, calculating the minimum space distance d min between the suspension arm and the wire based on the pseudo point cloud set and the suspension arm point cloud set.
7. The power line damage prevention monitoring system based on improved DepthPro depth estimation as set forth in claim 6, wherein the minimum spatial distance d min is expressed as follows: ; Wherein d min represents the minimum space distance between the suspension arm and the power transmission line, P arm represents the three-dimensional coordinate set of the key point of the suspension arm, P line represents the three-dimensional point set of the pseudo point cloud, and the symbol Representing Euclidean distance operations; for a point on the collection P arm , Corresponding to a point on P line .
8. The system for monitoring damage to a power transmission line based on improved DepthPro depth estimation as set forth in claim 7, wherein the risk assessment and classification early warning module includes: d1, adding uncertainty correction to minimum spatial distance d min : ; Wherein: Representing an estimated distance with a confidence interval; Representing the standard deviation corresponding to the uncertainty; d2, setting a risk level according to dest: And d3, triggering an audible and visual alarm and remote pushing when the risk level reaches a threshold value.
9. The system for monitoring damage to a power transmission line based on improved DepthPro depth estimation as set forth in claim 8, wherein d2, setting the risk level according to dest, comprises: When d est is more than 3m, judging that the safety is ensured; When 2m < d est is less than or equal to 3m, judging that the risk is mild; when 1m < d est is less than or equal to 2m, judging the risk as moderate; When d est is less than or equal to 1m, serious risk is judged.
10. The method for monitoring the damage to the power transmission line by using any one of the monitoring systems according to claims 1 to 9 is characterized by comprising the following steps: Step 1, defining an image IA which is an image of a crane boom and a power transmission line in a monitoring area at a time t, defining an image IB which is an image of the same area at a time t+1, and defining a risk label for training and verifying a monitoring model; step 2, performing target detection on the image IA and the image IB by using YOLOv network, identifying targets such as crane boom, lifting hook, tower, power transmission line and the like, and outputting a bounding box and a pixel level mask; Step 3, carrying out depth prediction on the image IA and the image IB by using an improved DepthPro model, wherein the model introduces a time sequence transducer module at the output end to model the cross-frame characteristic; step 4, generating a pseudo point cloud based on a detection result and a depth map, fitting and correcting a sag equation introduced into a power transmission line part, performing skeletonizing modeling on a crane boom, and extracting a key point set as a space representation of the boom; Step 5, calculating the minimum space distance between the suspension arm key points and the wire point set on the basis of the pseudo point cloud and the fitting wire curve to obtain a minimum space distance d min ; And 6, correcting the d min by combining an uncertainty propagation method to obtain an estimated distance d est with a confidence interval, classifying risks according to a threshold value, and triggering an acousto-optic alarm and a remote early warning if the risks exist.

Description

Power transmission line external damage prevention monitoring system and method based on improved DepthPro depth estimation Technical Field The invention relates to the technical field of power transmission line safety monitoring, in particular to a power transmission line external damage prevention monitoring system and method based on improved DepthPro depth estimation. Background The dynamic evolution of urban landscapes has profound effects on aspects such as urban planning, environmental monitoring, resource management and the like. Similarly, the problem of external damage to the transmission line is directly related to the safety and stability of the power system. In the construction process, the risk of collision between large construction machinery (such as a crane boom) and a power transmission line is continuously increased, and particularly, in a complex environment, how to accurately monitor the spatial relationship between the construction equipment and the power transmission line and timely perform early warning becomes a key problem of operation of a modern power system. With the continuous progress of the remote sensing technology, particularly the wide application of high-resolution satellites and unmanned aerial vehicles, the method can capture the changes of cities and surrounding areas with unprecedented precision and frequency, especially in the aspect of the interaction influence of buildings and power transmission lines. Currently, depth estimation techniques have made significant progress in the fields of image processing and computer vision, particularly in object recognition and scene understanding. However, depth estimation still faces many challenges in transmission line monitoring. Especially in construction sites, the relative positions of large-scale equipment such as crane booms and the like and a power transmission line need to be accurately captured and estimated. The conventional depth estimation methods mostly depend on a single frame image or a conventional sensor, but the methods are easily affected by illumination change, weather interference and distance change between targets in a complex construction environment, so that the depth estimation result is unstable and has larger error. The existing depth estimation method generally has the problems of insufficient global context information acquisition and insufficient time sequence dependent modeling, so that the depth prediction of a small target (such as an elongated power transmission line) and a complex structure (such as a crane boom) is inaccurate. The Chinese patent discloses a depth estimation network training method, a depth estimation method and an electronic device (application number: CN 202311718735.3), which provides a depth estimation network training method and aims to improve the image depth information precision based on monocular depth estimation, in particular to character depth estimation under a complex scene. According to the method, human semantic information and depth information are combined through combining a human analysis sub-network and a monocular depth estimation sub-network, so that the depth estimation network capable of better distinguishing each part of a person from the background is trained. In the training process, the loss weight is dynamically adjusted for different areas (such as a portrait and a background) in the training image so as to realize a more accurate depth estimation result. In addition, the depth estimation model further enables the depth estimation effect under night scenes and sports scenes to be more excellent through motion blur enhancement and dim light enhancement processing. Finally, the trained network can provide more accurate character depth information for blurring processing, and is particularly suitable for complex scenes such as night scenes, sports and the like. But limited by application range specificity, lack of adaptability to non-human targets (especially elongated structures), lack of timing modeling capabilities, and uniqueness of technical targets. The Chinese patent 'a light-weighted depth estimation method and system' (application number: 202510261430.7) discloses a light-weighted depth estimation method and system, and aims to solve the problems of large calculated amount and high memory consumption in the actual application of the existing binocular depth estimation method. According to the method, feature extraction and depth information processing are carried out on the binocular depth image by combining the two-dimensional depth separable convolution with the three-dimensional point-by-point convolution, so that multi-depth feature information is generated, and the calculation complexity and parameter quantity of the model are reduced. By carrying out depth estimation processing in the multilayer stacked hourglass network, the method can remarkably reduce calculation load and improve the practical application efficiency of the model