CN111079518B - Ground-falling abnormal behavior identification method based on law enforcement and case handling area scene - Google Patents
Ground-falling abnormal behavior identification method based on law enforcement and case handling area scene Download PDFInfo
- Publication number
- CN111079518B CN111079518B CN201911055068.9A CN201911055068A CN111079518B CN 111079518 B CN111079518 B CN 111079518B CN 201911055068 A CN201911055068 A CN 201911055068A CN 111079518 B CN111079518 B CN 111079518B
- Authority
- CN
- China
- Prior art keywords
- human body
- law enforcement
- network
- scene
- ground
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
Abstract
The invention belongs to the technical field of abnormal behavior analysis, and particularly relates to a method for identifying reverse abnormal behavior based on a scene of a law enforcement and case handling area, which can be divided into four parts of two-dimensional image data acquisition, human body detection based on deep learning, human body key point positioning and reverse behavior identification. Firstly, video image data in a law enforcement and case handling scene is obtained through a camera, (2) then the coordinate position of a human body detection frame is obtained through a target detection network, (3) then the position of each key point of a human body is obtained by using a human body key point detection network for each human body frame independently, (4) finally, each human body detection frame is input into a classification network for two classifications, whether the person falls down or not is judged, and therefore the recognition effect and recognition precision are improved.
Description
Technical Field
The invention belongs to the technical field of abnormal behavior analysis, and particularly relates to a land-fall abnormal behavior identification method based on a law enforcement and case handling area scene.
Background
The prior art comprises the following steps: aiming at the recognition method of the land-reversing behavior in the scene of law enforcement and case handling, a high recognition rate is needed so as to prevent life hazards of personnel in an active area due to land reversing. The existing judgment flow of the ground reversing behavior recognition is generally as follows: (1) acquisition of image data; (2) image data processing to obtain a human body image or a characteristic thereof; and (3) classifying the characteristics and identifying the ground reversing behavior.
The identification of the ground pouring behavior has different methods for the different steps. In the first stage, two-dimensional or three-dimensional image data of a human body can be obtained through a two-dimensional or three-dimensional camera; in the second stage, human body characteristics such as colors, length-width ratios of human body frames and the like can be manually obtained through a traditional method, and in addition, human body characteristics can be obtained through a deep learning method after human body frames are detected by a neural network and then automatic learning. And thirdly, performing two-classification on the extracted features through an SVM (support vector machine) or a deep learning classification network, and judging whether the extracted features are in a ground reversing behavior or not. The prior art can be further divided into a conventional method and a deep learning method according to whether the deep learning method is used or not. Compared with the method of the proposal, the problem of inaccurate identification can occur in the prior art when the human body detection frame obtained by deep learning is used for detecting the falling of the human body when the local human body appears.
For the existing traditional method for identifying the reverse behavior, the identification rate of the reverse behavior identification by using the manually extracted features is low, and the feature extraction is time-consuming and labor-consuming; the existing method for identifying the reverse abnormal behavior based on deep learning needs further enhancement in the identification effect although features are automatically learned and acquired. In order to further improve the recognition accuracy and meet the requirement of high recognition accuracy in actual projects, a new method for improving the recognition accuracy is required to be provided.
The present proposal proposes to use the identification of the ground-reversing behavior based on the key points of the human body. Unlike the prior art which uses Kinect (a three-dimensional motion sensing camera) to capture three-dimensional motion data of a human body in a three-dimensional space by means of a camera, the present proposal uses two-dimensional human body key point data acquired through an image to perform ground-reversing behavior recognition. And aiming at the scene of law enforcement and case handling, when a person is at the edge position of an image or is bent down to sit on the ground and only partial human bodies are detected, the prior art has the defects of easy false detection and incapability of accurately identifying the falling-to-ground behavior.
Disclosure of Invention
In order to solve the technical defects in the prior art, the invention designs a method for identifying the reverse abnormal behavior based on a scene of law enforcement and case handling.
The invention is realized by the following technical scheme:
a land-falling abnormal behavior identification method based on a law enforcement and case handling area scene comprises the following steps:
(1) Acquiring video image data in a law enforcement case handling scene through a camera;
(2) Obtaining the coordinate position of a human body detection frame through a target detection network;
(3) A human body key point detection network is independently used for each human body detection frame, and the position of each key point of the human body is obtained;
(4) And inputting each human body detection frame into a classification network to perform two classifications, and judging whether the person falls down.
Further, in step (1), the step of obtaining video image data in the law enforcement and case handling scene through the camera specifically includes obtaining, through an indoor camera installed in the law enforcement and case handling area, a moving video of a person in the camera shooting area, and obtaining image data including a human body based on a time sequence.
Further, in step (2), the obtaining the coordinate position of the human body detection frame through the target detection network specifically includes inputting the image data obtained in step (1) into the detection network based on the fast R-CNN algorithm to obtain the human body detection frame, and removing the redundant human body detection frame through NMS non-maximum inhibition, so that each human body has only a unique frame.
Further, in step (3), the step of using a human body key point detection network for each human body detection frame separately specifically includes inputting the human body detection block diagram obtained in step (2) into an SPPE network to obtain a human body key point distribution map.
Further, the positions of the key points include the head, neck, left and right shoulders, left and right elbows, left and right wrists, left and right hip joints, left and right knees and left and right ankles.
Further, in step (4), the step of determining whether the person falls down specifically includes the steps of:
4.1, inputting the coordinates of each human body detection frame and the coordinates of key points in each image;
4.2, judging the spatial relationship between the human body detection frame and the detection area, determining whether the human body is in the range set by the detection area, if so, executing the next step, otherwise, ending the identification;
4.3 judging whether the detected human body is a complete human body or not through the coordinate distribution of the key points of the human body, if so, inputting the detected human body into a two-class network, otherwise, ending the identification;
4.4, inputting the coordinate distribution diagram of the key points of the human body into a classification network to obtain a classification result;
and 4.5, judging whether the ground falling behavior is the ground falling behavior according to the classification result, if so, giving out a warning, otherwise, ending the identification.
Further, adding ResNet-18 composed of residual structures into the classification network with 18 layers of networks by the classification network in the step (4); wherein the residual structure function formula is:
Y＝F(x)+x
wherein x is the convolution characteristic of the input residual error structure; f (x) is a shortcut connection, and a convolution layer of a part surrounded by the residual structure is output; y is the output of the residual structure.
Further, the ResNet-18 network has an image input length and width of 224×224; the image sequentially passes through a convolution layer with 5 layers of convolution kernels of 3 multiplied by 3, a convolution layer with a characteristic dimension of 64, a convolution layer with 4 layers of convolution kernels of 3 multiplied by 3, a convolution layer with a characteristic dimension of 128, a convolution layer with 4 layers of convolution kernels of 3 multiplied by 3, a convolution layer with a characteristic dimension of 256, a convolution layer with 4 layers of convolution kernels of 3 multiplied by 3, and a convolution layer with a characteristic dimension of 512; inputting the data into a full convolution layer 2-d fc through an average pooling layer, wherein the full convolution layer 2-d fc carries out full convolution on the data with 512 dimensions into two dimensions; finally, a probability value of yes and no falling ground behaviors is obtained through a softmax (normalization operation) layer.
A computer readable storage medium having stored thereon a computer program, wherein the program when executed by a processor performs the steps of a method for identifying a ground fall anomaly in a law enforcement case handling scenario.
A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of a law enforcement-based ground fall anomaly behavior recognition method in a case of law enforcement.
Compared with the prior art, the invention has at least the following beneficial effects or advantages:
(1) By adding human body key point data in the identification process, part of human bodies are judged, false detection is effectively avoided, false detection rate is reduced, and the final ground falling behavior identification effect is improved;
(2) The inverse behavior recognition classification network uses a ResNet-18 network with a residual structure, so that recognition accuracy is improved.
Drawings
The invention will be described in further detail with reference to the accompanying drawings;
FIG. 1 is a general flow chart of the invention for ground-reversing behavior recognition;
FIG. 2 is a schematic diagram of key points of a human body;
FIG. 3 is a flowchart showing a method for identifying the reverse behavior;
FIG. 4 is a schematic diagram of a residual structure;
FIG. 5 is a diagram of a ResNet-18 network architecture.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The technical scheme provides a ground falling detection method based on key points of a human body, and an overall algorithm of the ground falling detection method comprises three parts: human body detection, human body key point positioning and classification model. According to the overall recognition steps, the method can be divided into four parts of two-dimensional image data acquisition, human body detection based on deep learning, human body key point positioning and ground reversing behavior recognition. Firstly, acquiring video image data in a law enforcement case-handling scene through a camera, (2) then obtaining the coordinate positions of human body detection frames through a target detection network, (3) then independently using a human body key point detection network for each human body frame to obtain the positions of each key point of a human body, and (4) finally inputting each human body detection frame into a classification network to perform two classifications, and judging whether the human body falls down or not, as shown in four general steps of fig. 1.
The detailed description of each step is as follows:
step 1, image data acquisition: and acquiring the activity video of the personnel in the shooting area by an indoor camera arranged in the law enforcement and case handling area to obtain image data containing the human body based on time sequence.
Step 2, human body detection: the method comprises the steps of inputting the image obtained in the step 1 into a detection network by using a fast R-CNN algorithm with higher instantaneity and accuracy in a target detection algorithm, finally obtaining a human body detection block diagram, and removing redundant human body detection block diagrams through NMS (non-maximum supression, non-maximal inhibition) so as to ensure that each human body has only a unique block.
Step 3, positioning key points of a human body: the obtained human body detection block image is input into an SPPE (single-person pose estimator) network to obtain a human body key point distribution map. Fig. 2 is a schematic diagram of 14 human body key points, wherein white solid circles are positions of the key points, including head, neck, left and right shoulders, left and right elbows, left and right wrists, left and right hip joints, left and right knees, and left and right ankles.
Step 4, identifying the reverse behavior: as shown in fig. 3, a flowchart for determining each human body frame is shown, and when a plurality of human body frames are determined, the input is circulated. Inputting each human body detection frame and key point coordinates thereof in each image, firstly judging the spatial relationship between the human body detection frame and a detection area, determining whether the human body is in the detection area, if so, executing the next step, otherwise, ending the identification; judging whether the detected human body is a complete human body or not through human body key point distribution, if so, inputting the detected human body into a two-class network, otherwise, ending the identification, and effectively solving the problem of false detection caused by inaccurate human body detection algorithm and only detecting local human body; inputting the human body key point coordinate distribution diagram into a two-class network to obtain a class result; and finally judging whether the ground falling action is performed according to the classification result, if so, giving a warning, otherwise, ending the identification.
In the step 4, the classification Network is used, and ResNet-18 (Residual Network-18 layer) formed by adding a Residual structure to the classification Network with 18 layers of networks is used, and fig. 4 is a schematic diagram of the Residual structure. By adding the residual error structure, the overall performance of the network is effectively improved, and the recognition effect is improved. Wherein the residual structure function formula is:
Y＝F(x)+x
wherein X is the convolution characteristic of the input residual error structure; f (x) is a shortcut connection (shortcut connections), and is output by a convolution layer of a part surrounded by a residual structure; y is the output of the residual structure.
The overall structure of the ResNet18 network is shown in FIG. 5, wherein the crossing arrow is a residual structure, the solid line is direct transfer, and the dotted arrow is up-scaled for keeping the output dimensions consistent; image (image) input length and width is 224×224;3×3conv,64 denotes a convolution kernel of 3×3, a feature dimension of 64, and the other layers are the same; the average pool layer performs dimension normalization on the images; 2-d fc represents a full convolution layer of two dimensions, and data of 512 dimensions are subjected to full convolution into the two dimensions; finally, a probability value of yes and no falling ground behaviors is obtained through a softmax (normalization operation) layer.
The present invention also provides a computer-readable storage medium having stored thereon a computer program, wherein the program when executed by a processor implements the steps of a method for identifying a fall-over anomaly behavior.
The invention also provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method for identifying the reverse ground abnormal behavior when executing the program.
The foregoing embodiments have been provided for the purpose of illustrating the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the foregoing embodiments are merely illustrative of the present invention and are not intended to limit the scope of the present invention. Any modification, equivalent replacement, improvement, etc. made without departing from the spirit and scope of the present invention also fall within the scope of the present invention.
Claims (9)
1. The method for identifying the reverse abnormal behavior based on the scene of law enforcement and case handling is characterized by comprising the following steps:
(1) Acquiring video image data in a law enforcement case handling scene through a camera;
(2) Obtaining the coordinate position of a human body detection frame through a target detection network;
(3) A human body key point detection network is independently used for each human body detection frame, and the position of each key point of the human body is obtained;
(4) Inputting each human body detection frame into a classification network to perform two classifications, and judging whether the person falls down;
in step (4), the step of judging whether the person falls down specifically includes the steps of:
4.1, inputting the coordinates of each human body detection frame and the coordinates of key points in each image;
4.2, judging the spatial relationship between the human body detection frame and the detection area, determining whether the human body is in the range set by the detection area, if so, executing the next step, otherwise, ending the identification;
4.3 judging whether the detected human body is a complete human body or not according to the coordinate distribution of the key points of the human body, if so, inputting the detected human body into a two-class network, otherwise, ending the identification;
4.4, inputting the coordinate distribution diagram of the key points of the human body into a classification network to obtain a classification result;
and 4.5, judging whether the ground falling behavior is the ground falling behavior according to the classification result, if so, giving out a warning, otherwise, ending the identification.
2. The method for recognizing the ground fall abnormal behavior based on the scene of law enforcement and case handling according to claim 1, wherein in the step (1), the step of obtaining the video image data in the scene of law enforcement and case handling by a camera specifically comprises obtaining the moving video of the person in the camera shooting area by an indoor camera installed in the area of law enforcement and case handling, and obtaining the image data containing the human body based on the time sequence.
3. The method for recognizing the ground fall abnormal behavior under the scene of law enforcement and case handling according to claim 1, wherein in the step (2), the step of obtaining the coordinate position of the human body detection frame through the target detection network specifically comprises the steps of inputting the image data obtained in the step (1) into the detection network based on a fast R-CNN algorithm to obtain the human body detection frame, and removing redundant human body detection frames through NMS non-great inhibition, so that each human body has only a unique frame.
4. The method for identifying the ground fall abnormal behavior based on the scene of law enforcement and case handling according to claim 1, wherein in the step (3), the step of using a human body key point detection network for each human body detection frame separately to obtain the position of each key point of the human body specifically comprises inputting the human body detection block diagram obtained from the step (2) into an SPPE network to obtain a human body key point distribution map.
5. The method for recognizing the abnormal behavior of the ground falling under the scene of law enforcement and case handling according to claim 4, wherein the positions of the key points comprise the head, the neck, the left and right shoulders, the left and right elbows, the left and right wrists, the left and right hip joints, the left and right knees and the left and right ankles.
6. The method for identifying the abnormal behavior of the ground fall based on the scene of law enforcement and case handling according to claim 1, wherein the classification network in the step (4) adds ResNet-18 composed of residual structures into the classification network with 18 layers of networks; wherein the residual structure function formula is:
Y＝F(x)+x
wherein x is the convolution characteristic of the input residual error structure; f (x) is a shortcut connection, and a convolution layer of a part surrounded by the residual structure is output; y is the output of the residual structure.
7. The method for identifying the reverse ground abnormal behavior based on the scene of law enforcement and case handling according to claim 6, wherein the image (image) input length and width of the ResNet-18 network is 224×224; the image sequentially passes through a convolution layer with 5 layers of convolution kernels of 3 multiplied by 3, a convolution layer with a characteristic dimension of 64, a convolution layer with 4 layers of convolution kernels of 3 multiplied by 3, a convolution layer with a characteristic dimension of 128, a convolution layer with 4 layers of convolution kernels of 3 multiplied by 3, a convolution layer with a characteristic dimension of 256, a convolution layer with 4 layers of convolution kernels of 3 multiplied by 3, and a convolution layer with a characteristic dimension of 512; inputting the data into a full convolution layer 2-d fc through an average pooling layer, wherein the full convolution layer 2-d fc carries out full convolution on the data with 512 dimensions into two dimensions; finally, a probability value of yes and no falling ground behaviors is obtained through a softmax (normalization operation) layer.
8. A computer readable storage medium having stored thereon a computer program, wherein the program when executed by a processor implements the steps of the method for identifying ground fall anomalies in a law enforcement-based case of any one of claims 1 to 7.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor performs the steps of the law enforcement-based abnormal behavior identification method of any of claims 1-7 when the program is executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911055068.9A CN111079518B (en) | 2019-10-31 | 2019-10-31 | Ground-falling abnormal behavior identification method based on law enforcement and case handling area scene |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911055068.9A CN111079518B (en) | 2019-10-31 | 2019-10-31 | Ground-falling abnormal behavior identification method based on law enforcement and case handling area scene |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111079518A CN111079518A (en) | 2020-04-28 |
| CN111079518B true CN111079518B (en) | 2023-08-29 |
Family
ID=70310638
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911055068.9A Active CN111079518B (en) | 2019-10-31 | 2019-10-31 | Ground-falling abnormal behavior identification method based on law enforcement and case handling area scene |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111079518B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111753747B (en) * | 2020-06-28 | 2023-11-24 | 高新兴科技集团股份有限公司 | Violent motion detection method based on monocular camera and three-dimensional attitude estimation |
| CN112255709A (en) * | 2020-10-21 | 2021-01-22 | 中铁第四勘察设计院集团有限公司 | Automatic lifting type flood monitoring device and method based on image recognition |
| CN113269105A (en) * | 2021-05-28 | 2021-08-17 | 西安交通大学 | Real-time faint detection method, device, equipment and medium in elevator scene |
| CN113361370B (en) * | 2021-06-02 | 2023-06-23 | 南京工业大学 | Abnormal behavior detection method based on deep learning |
| CN114419528B (en) * | 2022-04-01 | 2022-07-08 | 浙江口碑网络技术有限公司 | Anomaly identification method and device, computer equipment and computer readable storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107220604A (en) * | 2017-05-18 | 2017-09-29 | 清华大学深圳研究生院 | A kind of fall detection method based on video |
| CN110110613A (en) * | 2019-04-19 | 2019-08-09 | 北京航空航天大学 | A kind of rail traffic exception personnel's detection method based on action recognition |
| CN110287923A (en) * | 2019-06-29 | 2019-09-27 | 腾讯科技（深圳）有限公司 | Human body attitude acquisition methods, device, computer equipment and storage medium |
-
2019
- 2019-10-31 CN CN201911055068.9A patent/CN111079518B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107220604A (en) * | 2017-05-18 | 2017-09-29 | 清华大学深圳研究生院 | A kind of fall detection method based on video |
| CN110110613A (en) * | 2019-04-19 | 2019-08-09 | 北京航空航天大学 | A kind of rail traffic exception personnel's detection method based on action recognition |
| CN110287923A (en) * | 2019-06-29 | 2019-09-27 | 腾讯科技（深圳）有限公司 | Human body attitude acquisition methods, device, computer equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111079518A (en) | 2020-04-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111079518B (en) | Ground-falling abnormal behavior identification method based on law enforcement and case handling area scene | |
| CN110059558B (en) | Orchard obstacle real-time detection method based on improved SSD network | |
| CN108961235B (en) | Defective insulator identification method based on YOLOv3 network and particle filter algorithm | |
| CN113658132B (en) | Computer vision-based structural part weld joint detection method | |
| CN107657226B (en) | People number estimation method based on deep learning | |
| CN107330437B (en) | Feature extraction method based on convolutional neural network target real-time detection model | |
| WO2019104767A1 (en) | Fabric defect detection method based on deep convolutional neural network and visual saliency | |
| CN108416266B (en) | Method for rapidly identifying video behaviors by extracting moving object through optical flow | |
| CN106951870B (en) | Intelligent detection and early warning method for active visual attention of significant events of surveillance video | |
| CN111640157B (en) | Checkerboard corner detection method based on neural network and application thereof | |
| CN103699905B (en) | Method and device for positioning license plate | |
| WO2018165753A1 (en) | Structure defect detection using machine learning algorithms | |
| CN109859226B (en) | Detection method of checkerboard corner sub-pixels for graph segmentation | |
| CN105320917B (en) | A kind of pedestrian detection and tracking based on head-shoulder contour and BP neural network | |
| CN112288008B (en) | Mosaic multispectral image disguised target detection method based on deep learning | |
| CN105740945A (en) | People counting method based on video analysis | |
| CN113850242B (en) | Storage abnormal target detection method and system based on deep learning algorithm | |
| CN108229524A (en) | A kind of chimney and condensing tower detection method based on remote sensing images | |
| CN109389618A (en) | Foreground and background detection method | |
| CN113888461A (en) | Method, system and equipment for detecting defects of hardware parts based on deep learning | |
| CN109815887B (en) | Multi-agent cooperation-based face image classification method under complex illumination | |
| CN111160372A (en) | Large target identification method based on high-speed convolutional neural network | |
| CN103345762B (en) | Bayes's visual tracking method based on manifold learning | |
| CN109977892A (en) | Ship Detection based on local significant characteristics and CNN-SVM | |
| Wang et al. | Deep learning-based human activity analysis for aerial images |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |