Substation operation safety monitoring device

Abstract

The invention relates to the technical field of image processing, and discloses a transformer substation operation safety monitoring device which comprises a monitoring terminal, the monitoring terminal obtains collected video information of a main collection lens and auxiliary collection lenses, one main collection lens and a plurality of auxiliary collection lenses form a group, and the main collection lens and the auxiliary collection lenses form a group. Each group monitors a safety risk area, and the auxiliary acquisition lenses are distributed on the left side or the right side or the rear side of the main acquisition lens or the safety risk area; a video image collected by an auxiliary collection lens recommended by a lens region classification module is obtained through a fuzzy feature region selection module, after a fuzzy interest feature image is rapidly extracted, the image is transmitted to a feature analysis module for further feature judgment, and a secondary result obtained by the feature analysis module is supplemented with a fuzzy interest feature to obtain a fuzzy interest feature image. Therefore, the processing accuracy of the main acquisition lens is improved, and the processing amount is reduced.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a substation operation safety monitoring device. Background Art

[0002] In order to ensure the safe operation of the substation, the plant area will be covered by cameras for image acquisition. The complex camera system is built on the basis of obstacle avoidance acquisition, and the same location, especially key areas with frequent safety risks, will be captured by multiple cameras from multiple angles.

[0003] In order to improve the safety of employees' work in substations, image processing and deep learning are used to record and analyze the employees' postures in the collected images, determine the risk of their actions and provide prompts.

[0004] Since there are many cameras in the substation and a lot of video data is collected, if feature extraction and analysis are performed on multiple videos, a large amount of repeated feature extraction and analysis will be caused. If a single-area, single-lens acquisition method is used, due to the fixed video viewing angle, there will be errors in the motion capture of certain postures of the user in the acquisition lens, which can easily lead to misjudgment or omission of feature extraction. Given the high-risk characteristics of substations, there are certain monitoring risks. Summary of the invention

[0005] The object of the present invention is to provide a substation operation safety monitoring device to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A substation operation safety monitoring device comprises a monitoring terminal, wherein the monitoring terminal obtains collected video information of a main collection lens and a secondary collection lens, wherein one main collection lens and a plurality of secondary collection lenses form a group, and each group monitors a safety risk area, wherein the secondary collection lenses are distributed on the left side, right side or rear side of the main collection lens or the safety risk area;

[0008] The image information collected by the main acquisition lens is divided into multiple static frame images through a static frame acquisition module, and feature extraction and analysis are performed through a feature analysis module. The posture features of the personnel that cannot be accurately obtained by the feature analysis module are used as fuzzy interest features. The image partition grid module adds a partition grid to the image to quickly obtain the fuzzy interest feature coordinates, and the lens area classification module analyzes the fuzzy interest feature orientation and time of the main acquisition lens to select the secondary acquisition lens that can collect the fuzzy interest feature. The fuzzy feature area selection module obtains the video image in the secondary acquisition lens and extracts the static frame image in the corresponding time and corresponding grid area. The feature analysis module performs secondary analysis to obtain accurate feature output results.

[0009] As a further solution of the present invention: the fuzzy feature area selection module is replaced by a coordinate recording module and a coordinate conversion module is added, the coordinate recording module records the spatial coordinates of the fuzzy interest feature within the acquisition area of ​​the main acquisition lens, and after the lens area classification module selects the corresponding secondary acquisition lens, the coordinate conversion module obtains the coordinate information of the fuzzy interest feature in the image captured by the secondary acquisition lens through coordinate conversion based on the coordinates of the main acquisition lens, the secondary acquisition lens and the fuzzy interest feature coordinates obtained by the coordinate recording module. Then, the image corresponding to the fuzzy interest feature is quickly acquired.

[0010] As a further solution of the present invention: the video information collected by the auxiliary collection lens is in a normal collection and storage state on the basis that the main collection lens does not obtain the fuzzy interest feature.

[0011] As a further solution of the present invention: after the main acquisition lens obtains the fuzzy interest feature, the lens area classification module extracts the fuzzy interest feature map corresponding to the secondary acquisition lens in the same group at the same time point, and selects the unobstructed fuzzy interest feature map for the feature analysis module to perform feature extraction processing.

[0012] As a further solution of the present invention: the feature analysis module obtains the fuzzy interest features in the static frame acquisition module and also includes color on the basis of the posture features.

[0013] As a further solution of the present invention: a voice alarm module is also provided in the monitoring terminal, and the voice alarm module makes a voice prompt after the feature analysis module outputs the result.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] When the image information collected by the main acquisition lens cannot accurately obtain the action type after being processed by the feature analysis module, it becomes a fuzzy interest feature. Since the area collected by the main acquisition lens is fixed, the image collected by the main acquisition lens is framed through the image partition grid module, so as to quickly determine the frame area where the fuzzy interest feature is located. The lens area classification module cooperates with the frame area of ​​the fuzzy interest feature to quickly determine the secondary acquisition lens that can collect the frame area image of the fuzzy interest feature at other angles. The video image collected by the secondary acquisition lens recommended by the lens area classification module is obtained through the fuzzy feature area selection module, and the fuzzy interest feature image is quickly obtained according to the frame area. After the fuzzy interest feature image is quickly extracted, it is transmitted to the feature analysis module for further feature judgment. The secondary results obtained by the feature analysis module are supplemented with the fuzzy interest feature, and the data is recorded. A training set can be prepared later to further train the model.

[0016] The video information collected by the secondary acquisition lens is in a normal acquisition and storage state based on the fact that the main acquisition lens has not obtained the fuzzy interest features. Since the secondary acquisition lens is mainly used to monitor the environment under normal conditions, it does not need to analyze and process the actions of people in this state. The main function of the secondary acquisition lens is to provide other perspective references for auxiliary judgment based on the fact that the main acquisition lens cannot accurately judge the posture, thereby improving the accuracy of the main acquisition lens processing and reducing the processing volume. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0018] Figure 1 It is a schematic diagram of a first composition mode of a substation operation safety monitoring device;

[0019] Figure 2 It is a schematic diagram of the use of a substation operation safety monitoring device;

[0020] Figure 3 It is a schematic diagram of a substation operation safety monitoring device under two angles;

[0021] Figure 4 It is a schematic diagram of a second composition mode of a substation operation safety monitoring device;

[0022] In the figure: 1. Monitoring terminal; 2. Main acquisition lens; 3. Auxiliary acquisition lens; 4. Static frame acquisition module; 5. Feature analysis module; 6. Image partition grid module; 7. Lens area classification module; 8. Fuzzy feature area selection module; 9. Coordinate recording module; 10. Coordinate conversion module. DETAILED DESCRIPTION

[0023] See also Figure 1-Figure 4 : In this embodiment: it includes a monitoring terminal 1, which obtains the collected video information of the main collection lens 2 and the auxiliary collection lens 3, wherein one main collection lens 2 and multiple auxiliary collection lenses 3 form a group, and each group monitors a security risk area, wherein the auxiliary collection lenses 3 are distributed on the left side, right side or back side of the main collection lens 2 or the security risk area.

[0024] In this embodiment: the monitoring terminal 1 is mainly a general monitoring platform, and the collected information of multiple main acquisition lenses 2 and auxiliary acquisition lenses 3 are interconnected with the monitoring terminal 1 through wired or wireless means. Since the substation needs to achieve comprehensive lens coverage, the original intention of this coverage is not for image processing, but to ensure that the status in the substation is controlled by the monitoring terminal 1. In this case, there are many overlapping areas in the coverage of the acquisition lenses, especially for the number of monitoring lenses in the security risk area. On this basis, the main acquisition lens 2 and the auxiliary acquisition lens 3 are set, and the image information collected by the main acquisition lens 2 is analyzed by software after image processing and feature extraction of human movements and postures. However, the main acquisition lens 2 and the human perspective are relatively fixed, and some postures of the human body will overlap in the perspective of the main acquisition lens 2. For example, a curved line is curved on the X plane, and on the Y plane perpendicular to the X plane is a straight line composed of the endpoints of the curved line and the arc points of the curved line. Therefore, single-view analysis will produce certain omissions. Therefore, in order to solve this problem, a main acquisition lens 2 and a sub-acquisition lens 3 are set up to cooperate. The main acquisition lens 2 is used as the main object of image processing, and the fuzzy information collected by the main acquisition lens 2 is used to supplement the features through the sub-acquisition lens 3 to determine the output result, thereby improving the judgment accuracy and solving the problem of heavy workload caused by simultaneous feature processing of all lenses in the substation.

[0025] In this embodiment: The image information collected by the main acquisition lens 2 is segmented into multiple static frame images by the static frame acquisition module 4, and feature extraction and analysis are performed by the feature analysis module 5. The human posture features that cannot be accurately obtained by the feature analysis module 5 are used as fuzzy interest features. The image partitioning grid module 6 adds a partitioning grid to the image to quickly obtain the coordinates of the fuzzy interest features, and the lens area classification module 7 analyzes the orientation of the fuzzy interest features of the main acquisition lens 2 and selects the secondary acquisition lens 3 that can collect the fuzzy interest features according to the time. The fuzzy feature area selection module 8 obtains the video image within the secondary acquisition lens 3 and extracts the static frame images within the corresponding time and corresponding grid area, and performs secondary analysis through the feature analysis module 5 to obtain an accurate feature output result.

[0026] In this embodiment: After the image information collected by the main acquisition lens 2 is segmented into static frames by the static frame acquisition module 4, the feature analysis module 5 processes the features. First, the action is further decomposed into three channels: spatial, local time domain, and global time domain through three-channel CNNs. The static frame images segmented by the static frame acquisition module 4 are sent into the spatial channel. The local time domain uses optical flow features, and the optical flow algorithm is used to process the image. The optical flow is calculated in the three RGB color spaces respectively. The global time domain channel inputs the action difference image set, and the motion image frames, optical flow feature maps, and MSDI feature maps are input into the input layer of the CNNs. The input signal is connected to the convolutional layer and then to the pooling layer, and then convolution and pooling are repeated. Finally, one or two fully connected layers are connected. After defining the CNNs, first, the weights and parameters are randomly initialized to values that conform to the normal distribution, and the backpropagation algorithm is used to estimate each parameter in the framework, that is, by calculating the partial derivative of each neuron weight from back to front. Then, the preset weight is subtracted by the product of a learning rate operator and the partial derivative to update each weight. During the estimation, after the input set of the training set is learned by the model determined by the parameter set, an output set can be obtained. According to the least squares rule, the sum of the squared differences between the output set and the true value set is used as the cost function. The stochastic gradient descent method with mini-batch is used, that is, first, n training samples are selected, where n < m, and then n iterations are performed on these n samples, and only one sample is used each time. Then, the parameter is updated to the weighted mean of the parameters obtained from the n iterations. After iterating the above steps, the final estimation of the model parameters can be achieved. After obtaining the action depth features through the Three-stream CNNs framework, traditional CNNs-based action recognition methods often output them to a classification layer, that is, directly use softmax regression to classify each feature, and finally calculate the mean to obtain the type of the action.

[0027] In this embodiment: when the image information collected by the main acquisition lens 2 cannot accurately obtain the action type after being processed by the feature analysis module 5, it becomes a fuzzy interest feature. Since the area collected by the main acquisition lens 2 is fixed, the image collected by the main acquisition lens 2 is framed by the image partition grid module 6, so as to quickly determine the frame area where the fuzzy interest feature is located. The lens area classification module 7 cooperates with the frame area of ​​the fuzzy interest feature to quickly determine the auxiliary acquisition lens 3 that can collect the frame area image of the fuzzy interest feature at other angles. The fuzzy feature area selection module 8 obtains the video image collected by the auxiliary acquisition lens 3 recommended by the lens area classification module 7, and quickly obtains the fuzzy interest feature image according to the frame area. After the fuzzy interest feature image is quickly extracted, it is transmitted to the feature analysis module 5 for further feature judgment. The secondary result obtained by the feature analysis module 5 is supplemented with the fuzzy interest feature, and the data is recorded. A training set can be prepared later to further train the model.

[0028] In this embodiment: the fuzzy feature area selection module 8 is replaced by a coordinate recording module 9 and a coordinate conversion module 10 is added. The coordinate recording module 9 records the spatial coordinates of the fuzzy interest feature within the acquisition area of ​​the main acquisition lens 2. After the lens area classification module 7 selects the corresponding sub-acquisition lens 3, the coordinate conversion module 10 obtains the coordinate information of the fuzzy interest feature in the image captured by the sub-acquisition lens 3 through coordinate conversion based on the coordinates of the main acquisition lens 2, the sub-acquisition lens 3 and the fuzzy interest feature coordinates obtained by the coordinate recording module 9, and then quickly obtains the image corresponding to the fuzzy interest feature.

[0029] In this embodiment: first, the main acquisition lens 2 and the auxiliary acquisition lens 3 both establish a spatial coordinate system, the acquisition image of the main acquisition lens 2 is combined with the position of the fuzzy interest feature to calculate the spatial coordinates of the fuzzy interest feature relative to the main acquisition lens 2, and when the auxiliary acquisition lens 3 video is selected to capture the fuzzy interest feature, the coordinate system of the main acquisition lens 2 and the auxiliary acquisition lens 3 is converted through the coordinate conversion module 10, so that the fuzzy interest feature is converted from the spatial coordinates relative to the main acquisition lens 2 to the spatial acquisition coordinates relative to the auxiliary acquisition lens 3, so as to quickly locate it in the image captured by the auxiliary acquisition lens 3. This method is aimed at achieving rapid positioning when the coverage area of ​​the main acquisition lens 2 and the auxiliary acquisition lens 3 is greatly different.

[0030] In this embodiment: the video information collected by the auxiliary acquisition lens 3 is in a normal acquisition and storage state on the basis that the main acquisition lens 2 does not obtain the fuzzy interest feature. Since the auxiliary acquisition lens 3 is mainly used to monitor the environment under normal conditions, it is not necessary to analyze and process the actions of personnel in this state. The main function of the auxiliary acquisition lens 3 is to provide other perspective references for auxiliary judgment on the basis that the main acquisition lens 2 cannot accurately judge the posture, thereby improving the accuracy of the processing of the main acquisition lens 2 and reducing the processing volume.

[0031] In this embodiment: after the main acquisition lens 2 obtains the fuzzy interest feature, the lens area classification module 7 extracts the fuzzy interest feature map corresponding to the same group of secondary acquisition lenses 3 at the same time point, and selects the unobstructed fuzzy interest feature map for the feature analysis module 5 to perform feature extraction processing.

[0032] In this embodiment: the lens area classification module 7 extracts multiple auxiliary capture lens 3 videos that can capture the blurred interest feature area according to the time period, and segments the image through the static frame acquisition module 4. The image obtained after segmentation is further screened as an unobstructed image and is processed twice through the feature analysis module 5.

[0033] In this embodiment: the feature analysis module 5 obtains the fuzzy interest features in the static frame acquisition module 4 and also includes color on the basis of the posture features.

[0034] In this embodiment: if the color blurred interest feature obtained by the feature analysis module 5 is the only distinguishing point, there is no need to use the image partitioning grid module 6 for partitioning, or the coordinate recording module 9 and the coordinate conversion module 10 for positioning. It is only necessary to use color distinction to quickly obtain the blurred interest feature area captured by the secondary capture lens 3.

[0035] In this embodiment: a voice alarm module is also provided in the monitoring terminal 1, and the voice alarm module gives a voice prompt after the feature analysis module 5 outputs the result.

[0036] In this embodiment: when the final action is a risky action, a prompt can be given by voice to avoid security risks.

[0037] What is described above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A substation operation safety monitoring device, comprising a monitoring terminal (1), characterized in that: The monitoring terminal (1) obtains the collected video information of the main collection lens (2) and the auxiliary collection lens (3), wherein one of the main collection lenses (2) and a plurality of the auxiliary collection lenses (3) form a group, and each group monitors a security risk area, wherein the auxiliary collection lenses (3) are distributed on the left side, right side or rear side of the main collection lens (2) or the security risk area; The image information collected by the main acquisition lens (2) is divided into a plurality of static frame images by a static frame acquisition module (4), and feature extraction and analysis are performed by a feature analysis module (5). The posture features of the person that cannot be accurately obtained by the feature analysis module (5) are used as fuzzy interest features. The image partitioning grid module (6) adds a partitioning grid to the image to quickly obtain the coordinates of the fuzzy interest features. The lens area classification module (7) analyzes the fuzzy interest feature orientation and time of the main acquisition lens (2) to select the auxiliary acquisition lens (3) that can collect the fuzzy interest features. The fuzzy feature area selection module (8) obtains the video image in the auxiliary acquisition lens (3) and extracts the static frame image in the corresponding time and the corresponding grid area. The feature analysis module (5) performs secondary analysis to obtain accurate feature output results.

2. A substation operation safety monitoring device according to claim 1, characterized in that: The fuzzy feature area selection module (8) is replaced by a coordinate recording module (9) and a coordinate conversion module (10) is added. The coordinate recording module (9) records the spatial coordinates of the fuzzy interest feature within the acquisition area of ​​the main acquisition lens (2). After the lens area classification module (7) selects the corresponding auxiliary acquisition lens (3), the coordinate conversion module (10) obtains the coordinate information of the fuzzy interest feature in the image captured by the auxiliary acquisition lens (3) through coordinate conversion based on the coordinates of the main acquisition lens (2), the auxiliary acquisition lens (3) and the fuzzy interest feature coordinates obtained by the coordinate recording module (9), and then quickly obtains the image corresponding to the fuzzy interest feature.

3. A substation operation safety monitoring device according to claim 2, characterized in that: The video information collected by the auxiliary collection lens (3) is in a normal collection and storage state on the basis that the main collection lens (2) does not obtain fuzzy interest features.

4. A substation operation safety monitoring device according to claim 1, characterized in that: After the main acquisition lens (2) obtains the fuzzy interest feature, the lens area classification module (7) extracts the fuzzy interest feature map corresponding to the auxiliary acquisition lens (3) in the same group at the same time point, and selects the unobstructed fuzzy interest feature map for the feature analysis module (5) to perform feature extraction processing.

5. A substation operation safety monitoring device according to claim 1, characterized in that: The feature analysis module (5) obtains the fuzzy interest features in the static frame acquisition module (4) including color on the basis of the posture features.

6. A substation operation safety monitoring device according to claim 1, characterized in that: The monitoring terminal (1) is also provided with a voice alarm module, which provides a voice prompt after the feature analysis module (5) outputs a result.

Images