CN115966025A - Power operation abnormal behavior identification method based on human skeleton key points - Google Patents
Power operation abnormal behavior identification method based on human skeleton key points Download PDFInfo
- Publication number
- CN115966025A CN115966025A CN202310031814.0A CN202310031814A CN115966025A CN 115966025 A CN115966025 A CN 115966025A CN 202310031814 A CN202310031814 A CN 202310031814A CN 115966025 A CN115966025 A CN 115966025A
- Authority
- CN
- China
- Prior art keywords
- target
- frame
- space
- time
- human
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Abstract
The invention discloses a method for identifying abnormal behaviors of power operation based on key points of human skeletons, which comprises the following steps: step S1: acquiring a data set of abnormal behaviors of power operators; step S2: detecting target personnel on the video image by using a target detection algorithm, and tracking the detected target personnel by using a target tracking algorithm; and step S3: extracting human skeleton information from the target person detection frame in the step S2 by using an AlphaPose frame; and step S4: combining human body skeletons of target people in each frame of image of the video according to a time sequence to obtain a skeleton sequence, constructing a space-time diagram according to the skeleton sequence, extracting behavior characteristics of the space-time diagram by adopting space-time diagram convolution operation, classifying the behavior characteristics, and identifying whether abnormal behaviors exist in the target people. The invention realizes the accurate detection of the abnormal behavior and solves the problem that the abnormal behavior cannot be accurately identified in real time in the prior art.
Description
Technical Field
The invention relates to the technical field of video image processing and behavior recognition, in particular to a method for recognizing abnormal behaviors of power operation based on human skeleton key points.
Background
At present, the illegal behaviors such as no insulating gloves, no working clothes and the like in the power grid operation are inspected and supervised mainly by means of manual inspection, spot inspection and the like, an intelligent monitoring means of the operation behaviors is lacked, early warning is difficult to achieve, and a plurality of pain points such as large workload, easiness in negligence, poor instantaneity, low efficiency and the like exist. The operation workers have the disadvantages of poor safety production awareness, untight safety operation management, severe habitual violation and potential safety hazards easily caused in operation. In the aspect of inspection and supervision, a manager cannot monitor and guide the field operating personnel, and tracking inspection and supervision are lacked. With the development of artificial intelligence technology in recent years, intelligent assessment of safety state of power fields based on visual analysis has become possible. However, most of the current research is manual operation on scene identification and data processing, the processing efficiency of the scene is low, and real-time monitoring cannot be carried out. Meanwhile, in practical application, the problems that the monitoring visual angle changes, the postures of pedestrians change, the detection objects are easily shielded and the like exist, and the reliability of electric power operation safety behavior management and control application is influenced.
Due to the fact that the monitoring scene has the conditions of strong and weak light, monitoring angle, blocking of pedestrians by objects, complex background of the monitoring scene and the like, the traditional neural network identification facing to the pedestrian attribute identification has the problems of low operation efficiency, inaccurate identification precision and the like. Meanwhile, due to the limited conditions of the operation place, the actions of the operators are complex, so that the application of pedestrian attribute identification in the aspect of electric power safety operation behavior control is difficult, and the real-time monitoring and the overall control on the safety condition of the whole construction site cannot be realized; the problems of potential safety hazards, non-standard operation and the like are found and difficult to feed back and remind in real time.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a method for identifying abnormal behaviors of power operation based on human skeleton key points, which solves the problem of monitoring unsafe behaviors of power operation personnel without wearing safety helmets, wearing protective clothing and the like in real time, realizes intelligent real-time identification and intelligent early warning of operation violation behaviors, has the functions of real-time reminding, remote control and the like of behaviors with potential safety hazards and ensures the production safety.
In order to achieve the purpose of the invention, the invention adopts the technical scheme that: a method for recognizing abnormal behaviors of power operation based on key points of human bones is disclosed. The method comprises the following specific steps:
step S1: acquiring a data set of abnormal behaviors of power operators;
step S2: detecting target personnel on the video image by using a target detection algorithm, and tracking the detected target personnel by using a target tracking algorithm;
and step S3: extracting human skeleton information from the target person detection frame in the step S2 by using an AlphaPose frame;
and step S4: the captured video images are processed frame by frame, human body skeleton models of workers are constructed and are arranged according to the time sequence to obtain a skeleton sequence, a space-time diagram is established according to the obtained skeleton sequence, the space-time diagram is subjected to convolution operation, so that the behavior characteristics of the target workers can be obtained, the obtained behavior characteristics are classified, and finally whether the operation behavior is abnormal or not can be judged.
Further, the specific steps in step S2 are:
step S21: and carrying out personnel detection on the image by using a YOLOv5 target detection algorithm fused with CSPNet to obtain a target personnel boundary frame. Two structures of cross-phase local networks (CSPNet) are adopted on the network main branch, so that the learning capacity of the convolutional neural network is enhanced, the calculation bottleneck is eliminated, and the memory cost is reduced. At the input end of the neural network, the method adopts a Mosaic data enhancement mode, and splices the pictures in a random scaling, random cutting and random arrangement mode, so that the detection effect on small objects is enhanced. A Focus structure is adopted in a BackBone part of the neural network, a CSPNet _1 structure is adopted, and a cross-stage method is adopted, so that gradient paths of the BackBone part are increased. Meanwhile, a CSPNet structure different from the BackBone part is adopted in the Neck part, and only ordinary convolution operation is applied, so that the fusion of characteristics is enhanced.
Step S22: the deep sort algorithm is adopted for multi-target tracking. Deepsiprt is a current excellent target tracking algorithm, kalman filtering and Hungarian algorithms are adopted to match the motion trail of a target, besides the basic function of the Sort algorithm, deepsiprt also introduces the calculation of appearance characteristics, the precision and the speed are improved, and the problem of ID jump is effectively solved. Importing the target person information detected by YOLOv5 into a Deepsort frame, and performing cascade matching on each frame of information of the target image and the previous frame so as to obtain a target person tracking model with higher accuracy;
further, the specific steps in step S3 are:
step S31: alphapose proposes a regional multi-person pose estimation (RMPE) framework. The RMPE framework consists of three new components: symmetric STN with parallel SPPE, parametric pose NMS and pose guidance suggestion generator (PGPG);
step S32: PGPG studies training data in large quantities by learning the conditional distribution suggested by the bounding box of a given human posture. Because of the use of symmetric STNs and parallel SPPEs, SPPEs are beginning to accept processing of human localization errors;
step S33: the last redundant detection can be cut down using the parameter posture NMS.
Further, the specific steps of step S4 are:
step S41: a space-time diagram G = (V, E) was constructed on a T-frame skeleton sequence containing N joints. All joints in a skeletal sequence are represented by a set of nodes V = { V = } ti I T =1, \\8230;, T, i =1, \8230;, N } contains, the edge set E includes two subsets, one is a spatial edge set E used to describe the connection of the internal skeleton within each frame S ＝{v ti v tj L (i, j) E H, where H represents a set of naturally connected human joints, and the other is a time edge E describing the connection of the same joint in successive frames F ＝{v ti v (t+1)i }。
Step S42: the specific measures for judging whether the target person has abnormal behaviors are as follows by carrying out convolution operation on the space diagram, generating a characteristic diagram with a higher level on the diagram and classifying the characteristic diagram: regularizing a node set, a space edge, a time edge and the like, and sending the regularized node set, the space edge, the time edge and the like into a space-time graph convolution network; the space-time diagram convolution network comprises nine layers of space-time diagram convolution, 64 channels are output by the first three layers, 128 channels are output by the middle three layers, 256 channels are output by the last three layers, 9 time convolution kernels are totally arranged, each space-time diagram convolution layer is connected by using residual errors, and the time convolution layers of the 4 th layer and the 7 th layer are set as posing layers; and carrying out global posing on the output of the 256 channels to obtain 256-dimensional feature vectors, classifying by softmax, and identifying whether the power operation personnel have operation abnormal behaviors according to the classification result.
The invention has the beneficial effects that:
1. the YOLOv5 target detection framework fused with CSPNet is used for eliminating the computation bottleneck, reducing the memory cost and strengthening the fusion of the characteristics. The accuracy of target detection is improved.
2. The method combines a Deepsort algorithm and an improved YOLOv5 framework, and effectively realizes high-performance real-time multi-target tracking of the power operators by setting a basic threshold parameter to control the Euclidean distance of track prediction.
3. The method has the advantages that the detected target personnel are subjected to skeleton extraction by using the alpha position deep learning network, whether abnormal behaviors exist in the target personnel is identified by using the space-time graph convolution network, the abnormal behaviors are accurately detected, and the problem that the abnormal behaviors cannot be accurately identified in real time in the prior art is solved.
Drawings
Fig. 1 is a schematic flow chart of a method for identifying abnormal behavior of power operation based on key points of human bones according to an embodiment of the present invention.
FIG. 2 is a diagram of the CSPNet _1 neural network structure of the present invention.
FIG. 3 is a schematic diagram of a CSPNet-YOLOv5 model network according to the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
As shown in fig. 1, the method for identifying abnormal behavior of power operation based on key points of human skeleton of the present invention specifically includes the following steps:
step S1: acquiring a data set of abnormal behaviors of power operators;
the training data set is obtained through a manual acquisition method, namely a group of videos in the power operation environment are shot, and then a plurality of video frame images are captured from the videos. The video frame images are processed by image enhancement and the like to obtain more samples, so that the data generated by the training data set is very close to the original video frame images. This allows the scale of the training data set to be increased. The generalization capability of the model is improved under the treatment of image augmentation, so that the dependence degree of the model on certain specific attributes is reduced. The expansion of the data set is realized by turning the image left and right, translating and the like.
Step S2: firstly, target person detection is carried out on a video image by using a target detection algorithm, and then the detected target person is tracked by using a target tracking algorithm;
the method uses the YOLOv5 network fused with CSPNet to detect the human body in the video, and adopts the deep sort algorithm to perform multi-target tracking. The specific implementation is as follows:
step S21: and carrying out personnel detection on the image by utilizing a YOLOv5 target detection algorithm fused with CSPNet to obtain a target personnel boundary frame. Two structures of cross-phase local networks (CSPNet) are adopted on the network main branch, so that the learning capability of the convolutional neural network is enhanced, the calculation bottleneck is eliminated, and the memory cost is reduced. At the input end of the neural network, the method adopts a Mosaic data enhancement mode, and splices pictures in a random scaling, random cutting and random arrangement mode, so that the detection effect on small objects is enhanced. The Focus structure is adopted in the BackBone part of the neural network, and the CSPNet _1 structure is adopted, and the structure is shown in FIG. 2. And a cross-phase method is adopted, so that the gradient path of the BackBone part is increased. The invention also adopts a CSPNet structure different from the BackBone part in the Neck part, only applies common convolution operation and strengthens the fusion of the characteristics.
The electric power operating personnel are important control objects on the electric power operating site, and real-time accurate detection is the core of safety control. The fused CSPNet-YOLOv5 model provided by the invention is applied to monitoring of power operation behaviors. The fusion model adopts the full-image information for prediction, and the full-image information is utilized in the training and prediction processes. The principle of the detection model is shown in fig. 3.
Step S22: the DeepsSort algorithm is adopted for multi-target tracking. Deepsort is a relatively advanced target tracking algorithm at present, target track matching is carried out by adopting Kalman filtering and Hungarian algorithm, and appearance characteristic calculation is introduced on the basis of the Sort algorithm, so that the problem of ID jump is solved, and the method has higher precision and speed. Importing the detection frame information of the target person identified by YOLOv5 into a Deepsort framework; and performing cascade matching on the target information of the frame and the track information of the previous frame to finally obtain a target tracking model capable of accurately tracking the power operation personnel.
And step S3: extracting human skeleton information from the target person detection frame in the step S2 by using an AlphaPose frame;
there are two mainstream methods for estimating the posture of a human body at present: two-step frames and Part-based frames. Two-step frame detects each target human body detection frame in the target environment first, and then detects the posture of a single human body area (top-down method) respectively. The method of obtaining a multi-person skeleton (bottom-up method) by splicing each of the body nodes obtained by environment detection has a drawback that if the distance between two target detection persons is too close, it is possible to generate ambiguity, and in addition, because of the dependence on the relationship between two parts, the global information is not sufficiently obtained.
AlphaPose belongs to a top-down approach, proposing an RMPE (regional multi-person pose detection) framework. The framework mainly comprises Symmetry Spatial Transform Network (SSTN), parametric Pose Non-Maximum-compression (NMS) and Pose-Guided probes Generator (PGPG). And three technologies of Symmetry Spatial Transform Network (SSTN), deep pro-posals generator (DPG) and parameter posi-nonmaximum suppression (p-NMS) are used for solving the multi-person attitude estimation problem in the field scene. The addition of SSTN to SPPE structures can result in high quality body regions in a less precise range. Parallel SPPE branches (SSTNs) optimize their own network. By adopting a self-defined attitude distance measuring method, the similarity between different attitudes is compared. The attitude distance parameters are optimized by a data-based method. And finally, enhancing the training data by utilizing a PGPG algorithm, simulating the generation of a human body region frame, and learning the description of each posture on the basis of the generation of the human body region frame, thereby obtaining a larger training set. PGPG studies training data in large quantities by learning a conditional distribution suggested by a boundary box of a given human body posture. Because of the use of symmetric STNs and parallel SPPEs, SPPEs are beginning to accept processing of human localization errors. The last redundancy check may be subtracted using the parameter posture NMS.
And step S4: the captured video images are processed frame by frame, a human body skeleton model of a worker is constructed, the human body skeleton model and the human body skeleton model are arranged according to the time sequence to obtain a skeleton sequence, a space-time diagram is established according to the obtained skeleton sequence, and space-time diagram convolution (ST-GCN) operation is carried out on the space-time diagram to obtain the behavior characteristics of the target worker, the obtained behavior characteristics are classified, and whether abnormity exists during traveling can be judged finally.
Step S41: and constructing a graph based on the key points of the human body.
We constructed an undirected space-time diagram G = (V, E) over a skeletal sequence that contains N joints and T frames with intra-and inter-frame connections, i.e., a spatio-temporal graph.
The combination method comprises the following steps: taking each human body key point in a frame as a node, and taking natural connection and time domain connection between the human body key points as edge to form a graph (which can be understood as a three-dimensional graph)
Construct a single frame graph (spatial domain): the same node is found in consecutive frames and concatenated into time domain information (information of edge). The edge information consists of two subsets.
The first subset is: relationships between keypoints in the same frame (i denotes different keypoints in the same frame, j denotes the same keypoint between different frames)
E S ＝{v ti v tj ∣(i,j)∈H}
A second subset of: relationship between human body joints between different frames
E F ＝{v ti v (t+1)i }
So far, a graph containing spatial and temporal information is formed according to the human key point information of different frames in a video.
Wherein the node information is (abscissa, ordinate, confidence); the side information is (ES, EF).
Construct inter-frame graph (time domain): the same node is found in consecutive frames and concatenated into time domain information (information of edge). The edge information consists of two subsets.
The first subset is: relationships between keypoints in the same frame (i denotes different keypoints in the same frame, j denotes the same keypoint between different frames)
E S ＝{v ti v tj ∣(i,h)∈H}
A second subset of: relationship between human body joints between different frames
E F ＝{v ti v (t+1)i }
So far, a graph containing spatial and temporal information is formed according to the human key point information of different frames in a video.
Wherein the node information is (abscissa, ordinate, confidence); the side information is (ES, EF).
In summary, the invention firstly detects the target by using the YOLOv5 network fused with the CSPNet, and then performs continuous tracking by combining the deep sort algorithm; further extraction of human body skeleton characteristics is achieved by adopting an AlphaPose attitude estimation framework, and finally, a space-time graph convolution network is used for classifying skeleton sequences, so that identification and alarm of abnormal operation behaviors of power operators are achieved.
The foregoing merely represents a preferred embodiment of the invention, which is described in some detail and detail, and is not to be construed as limiting the scope of the invention. It should be noted that, for those skilled in the art, various changes, modifications and substitutions can be made without departing from the spirit of the present invention, and these are all within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (4)
1. A method for identifying abnormal behaviors of power operation based on key points of human bones is characterized by comprising the following steps: the method comprises the following steps:
step S1: acquiring a data set of abnormal behaviors of power operators;
step S2: detecting target personnel on the video image by using a target detection algorithm, and tracking the detected target personnel by using a target tracking algorithm;
and step S3: extracting human skeleton information from the target person detection frame in the step S2 by using an AlphaPose frame;
and step S4: the captured video images are processed frame by frame, a human body skeleton model of a worker is constructed, the human body skeleton model and the human body skeleton model are arranged according to the time sequence to obtain a skeleton sequence, a space-time diagram is established according to the obtained skeleton sequence, the space-time diagram is subjected to convolution operation to obtain the behavior characteristics of a target worker, the obtained behavior characteristics are classified, and whether the operation behavior meets the standard or not is judged.
2. The method for recognizing the abnormal behavior of the power operation based on the key points of the human skeleton as claimed in claim 1, wherein the method comprises the following steps: the specific steps in the step S2 are as follows:
step S21: carrying out personnel detection on the image by utilizing a YOLOv5 target detection algorithm fused with CSPNet to obtain a target personnel boundary frame; two structures of cross-phase local networks are adopted on a network master; splicing the pictures at the input end of the neural network in a mode of random zooming, random cutting and random arrangement by adopting a Mosaic data enhancement mode; a Focus structure is adopted in a BackBone part of the neural network, a CSPNet _1 structure is adopted, and a cross-stage method is adopted to increase a gradient path of the BackBone part; meanwhile, a CSPNet structure different from that of the BackBone part is adopted in the Neck part, and only ordinary convolution operation is applied to enhance the fusion of characteristics;
step S22: performing multi-target tracking by adopting a DeepSort algorithm; and importing the target person information detected by YOLOv5 into a Deepsort frame, and performing cascade matching on each frame of information of the target image and the previous frame to finally obtain a target tracking model capable of accurately tracking the power operation person.
3. The method for identifying abnormal behaviors of power operation based on key points of human bones as claimed in claim 1, wherein the method comprises the following steps: the specific steps in the step S3 are as follows:
step S31: alphapos proposes a regional multi-person attitude estimation RMPE framework; the RMPE framework consists of three new components: symmetrical STN with parallel SPPE, parameter posture NMS and posture guidance suggestion generator PGPG;
step S32: the PGPG learns the condition distribution of the boundary box suggestion of the preset human body posture so as to demonstrate a large amount of training data; because of the use of symmetric STNs and parallel SPPEs, SPPEs are beginning to accept processing of human localization errors;
step S33: the parameter posture NMS is used to reduce redundant detection.
4. The method for identifying abnormal behaviors of power operation based on key points of human bones as claimed in claim 1, wherein the method comprises the following steps: the specific steps in the step S4 are as follows:
step S41: constructing a space-time diagram G = (V, E) on a T-frame skeleton sequence containing N joints; all joints in a skeletal sequence are represented by a set of nodes V = { V = } ti I T =1, \\8230;, T, i =1, \8230;, N } contains, the edge set E includes two subsets, one is a spatial edge set E used to describe the connection of the internal skeleton within each frame S ＝{v ti v tj L (i, j) E H, where H represents a set of naturally connected human joints, and the other is a time edge E describing the connection of the same joint in successive frames F ＝{v ti v (t+1)i }；
Step S42: the specific measures for judging whether the target person has abnormal behaviors are as follows by carrying out convolution operation on the space diagram, generating a characteristic diagram with a higher level on the diagram and classifying the characteristic diagram: regularizing a node set, a space edge and a time edge, and sending the regularized node set, the space edge and the time edge into a space-time graph convolution network; the space-time diagram convolution network comprises nine layers of space-time diagram convolution, 64 channels are output by the first three layers, 128 channels are output by the middle three layers, 256 channels are output by the last three layers, 9 time convolution kernels are totally arranged, each space-time diagram convolution layer is connected by using residual errors, and the time convolution layers of the 4 th layer and the 7 th layer are set as posing layers; and carrying out global posing on the output of the 256 channels to obtain 256-dimensional feature vectors, classifying by softmax, and identifying whether the power operation personnel have operation abnormal behaviors according to the classification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310031814.0A CN115966025A (en) | 2023-01-10 | 2023-01-10 | Power operation abnormal behavior identification method based on human skeleton key points |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310031814.0A CN115966025A (en) | 2023-01-10 | 2023-01-10 | Power operation abnormal behavior identification method based on human skeleton key points |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115966025A true CN115966025A (en) | 2023-04-14 |
Family
ID=87363364
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310031814.0A Pending CN115966025A (en) | 2023-01-10 | 2023-01-10 | Power operation abnormal behavior identification method based on human skeleton key points |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115966025A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116959099A (en) * | 2023-06-20 | 2023-10-27 | 河北华网计算机技术有限公司 | Abnormal behavior identification method based on space-time diagram convolutional neural network |
-
2023
- 2023-01-10 CN CN202310031814.0A patent/CN115966025A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116959099A (en) * | 2023-06-20 | 2023-10-27 | 河北华网计算机技术有限公司 | Abnormal behavior identification method based on space-time diagram convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | DP-SLAM: A visual SLAM with moving probability towards dynamic environments | |
| Wang et al. | Interior attention-aware network for infrared small target detection | |
| Coşar et al. | Toward abnormal trajectory and event detection in video surveillance | |
| Gu et al. | Multi-person hierarchical 3d pose estimation in natural videos | |
| CN112200106A (en) | Cross-camera pedestrian re-identification and tracking method | |
| Han et al. | Laneformer: Object-aware row-column transformers for lane detection | |
| Chen et al. | Multitarget tracking in nonoverlapping cameras using a reference set | |
| Win et al. | Real-time human motion detection, tracking and activity recognition with skeletal model | |
| Lin et al. | How generative adversarial networks promote the development of intelligent transportation systems: A survey | |
| Liang et al. | Methods of moving target detection and behavior recognition in intelligent vision monitoring. | |
| Hegde et al. | Uncertainty-aware mean teacher for source-free unsupervised domain adaptive 3d object detection | |
| CN115966025A (en) | Power operation abnormal behavior identification method based on human skeleton key points | |
| Divya et al. | Inspection of suspicious human activity in the crowdsourced areas captured in surveillance cameras | |
| Shi et al. | An underground abnormal behavior recognition method based on an optimized alphapose-st-gcn | |
| Khatri et al. | Detection of animals in thermal imagery for surveillance using GAN and object detection framework | |
| CN114373225A (en) | Behavior recognition method and system based on human skeleton | |
| Hegde et al. | Source-free unsupervised domain adaptation for 3d object detection in adverse weather | |
| Ye et al. | An efficient few-shot object detection method for railway intrusion via fine-tune approach and contrastive learning | |
| Zhu et al. | Dualda-net: Dual-head rectification for cross domain object detection of remote sensing | |
| Fang et al. | Vision-based traffic accident detection and anticipation: A survey | |
| CN113076825A (en) | Transformer substation worker climbing safety monitoring method | |
| Ichihara et al. | Accuracy evaluations of video anomaly detection using human pose estimation | |
| Naeem et al. | Vision-Based Real-time Human Malicious Behavior Detection | |
| Bhargava et al. | A comprehensive study and detection of anomalies for autonomous video surveillance using neuromorphic computing and self learning algorithm | |
| Jin et al. | Research on Human Action Recognition Based on Global-Local Features of Video |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |