CN111144229A - Posture detection system and method - Google Patents
Posture detection system and method Download PDFInfo
- Publication number
- CN111144229A CN111144229A CN201911232459.3A CN201911232459A CN111144229A CN 111144229 A CN111144229 A CN 111144229A CN 201911232459 A CN201911232459 A CN 201911232459A CN 111144229 A CN111144229 A CN 111144229A
- Authority
- CN
- China
- Prior art keywords
- image
- detection module
- detection
- module
- attitude detection
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 128
- 238000000034 method Methods 0.000 title claims abstract description 23
- 238000012937 correction Methods 0.000 claims abstract description 20
- 238000003702 image correction Methods 0.000 claims abstract description 13
- 238000013528 artificial neural network Methods 0.000 claims description 12
- 230000036544 posture Effects 0.000 abstract description 56
- 238000012545 processing Methods 0.000 abstract description 6
- 239000013589 supplement Substances 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 239000000047 product Substances 0.000 description 2
- 208000028752 abnormal posture Diseases 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000001154 acute effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 210000003127 knee Anatomy 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000032258 transport Effects 0.000 description 1
- 210000000689 upper leg Anatomy 0.000 description 1
Images
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
- 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
-
- 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
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/242—Aligning, centring, orientation detection or correction of the image by image rotation, e.g. by 90 degrees
-
- 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
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The application belongs to the technical field of image processing, and particularly relates to a posture detection system and method. The existing gesture detection methods are also easy to be subjected to human body postures of different angles and different gestures on component combination, so that the recognition accuracy is not high. The utility model provides an attitude detection system, including image acquisition unit, image detection unit and the result output unit who connects gradually, image detection unit is including half body attitude detection module, image correction module and the whole body attitude detection module that connects gradually, half body attitude detection module with image acquisition unit connects, half body attitude detection module with whole body attitude detection module connects, whole body attitude detection module with the result output unit connects. The gesture recognition is more accurate, and meanwhile, the gesture correction link effectively supplements the phenomenon of inaccurate detection results caused by various gestures under the common condition.
Description
Technical Field
The application belongs to the technical field of image processing, and particularly relates to a posture detection system and method.
Background
At present, computer vision methods based on deep learning have been widely applied to various industrial products. Attitude detection is also known as attitude estimation. Namely, the pedestrian pictures or videos in a conventional image are labeled by means of a computer technology. For example, the head, shoulder, elbow and knee of a human body are marked. The posture detection method can detect 18 joint points of a person in a video or an image, and returns high-precision coordinates of the joint points to realize drawing of the personnel posture. The technology can logically judge the behavior and the action of the character through the analysis of data of each joint point, thereby knowing what the current character is doing or in what state.
The gesture detection method can distinguish whether specific operations are legal or not by analyzing the human body gesture, such as operations of carrying chemical products, operations of installing important parts and the like. The gesture detection method can detect the occurrence of the alarm in time, such as walking and shaking, or no movement for a long time and the like, by analyzing the personnel captured by the video.
In the existing posture detection methods, insufficient calculation force exists in front of a large number of human body parts to be detected, so that the calculation time is increased. Meanwhile, the component combination is also easy to be subjected to human body states with different angles and different postures, so that the recognition accuracy is not high; some attitude detection methods have a greatly reduced detection effect in the face of abnormal attitudes such as non-upright attitude; there are also some video processing times that are long and require the user to wait.
Disclosure of Invention
1. Technical problem to be solved
Based on the existing posture detection methods, insufficient calculation force can occur in front of a large number of human body parts to be detected, so that the calculation time is increased. Meanwhile, the component combination is also easy to be subjected to human body states with different angles and different postures, so that the recognition accuracy is not high; some attitude detection methods have a greatly reduced detection effect in the face of abnormal attitudes such as non-upright attitude; there are also some problems that the video processing time is long and the user is required to wait, and the application provides a posture detection system and method.
2. Technical scheme
In order to achieve the above object, the present application provides an attitude detection system including an image acquisition unit, an image detection unit, and a result output unit that are connected in sequence, the image detection unit includes a half-body attitude detection module, an image correction module, and a whole-body attitude detection module that are connected in sequence, the half-body attitude detection module is connected to the image acquisition unit, the half-body attitude detection module is connected to the whole-body attitude detection module, and the whole-body attitude detection module is connected to the result output unit.
Another embodiment provided by the present application is: the half-body posture detection module is a light-weight posture detection module and is used for detecting the head, the two shoulders and the two crotches.
Another embodiment provided by the present application is: the half-body posture detection module comprises a plurality of layers of neural networks.
Another embodiment provided by the present application is: the image rectification module includes a rotation sub-module.
Another embodiment provided by the present application is: the half-body posture detection module comprises an image correction judgment submodule.
The application also provides a posture detection method, which comprises the following steps:
step 1: acquiring an image, and detecting the half-body posture of the image;
step 2: determining whether the image requires rectification;
and step 3: detecting the posture of the whole body of the image which does not need to be corrected and outputting a detection result; and (4) carrying out image correction on the image to be corrected, carrying out whole body posture detection, and outputting a detection result.
Another embodiment provided by the present application is: in the step 2, if the portrait is perpendicular to the image, no correction is needed, and if the portrait is not perpendicular to the image, the correction is needed.
Another embodiment provided by the present application is: and in the step 3, the image is rectified into a rotation image so that the portrait is vertical to the image.
Another embodiment provided by the present application is: the image rectification is realized through an API interface.
Another embodiment provided by the present application is: the attitude detection method is based on a neural network basic computing platform, and the neural network computing platform is deployed on a TensorFlow open source framework with a distributed function.
3. Advantageous effects
Compared with the prior art, the gesture detection system and method provided by the application have the beneficial effects that:
the posture detection system provided by the application discovers the relation between the human posture position and the final accuracy rate based on production practice, and provides a reference line method for assisting the image and correcting the position of the pedestrian in the image. The corrected image can realize more accurate posture recognition, and meanwhile, the posture correction link effectively supplements the phenomenon of inaccurate detection results caused by various postures under the common condition.
The posture detection system provided by the application skillfully embeds the posture correction algorithm into the posture detection algorithm, and designs two detection modules, one large detection module and one small detection module, so as to finish accurate human body posture detection. The pressure of the whole detection network is effectively relieved by one large network and one small network, the existing resources are fully utilized after the parallelization, and the input is ensured not to stop the module and not to be interrupted.
The posture detection method provided by the application is based on a general computing platform, two networks, namely one large network and one small network, are deployed on a basic platform in a parallelization mode, and the human body posture detection in the video is realized quickly. Compared with the existing serial solution, the parallel method based on the general computing platform is fully matched with the algorithm in the application, so that the video attitude detection time is greatly shortened.
Drawings
FIG. 1 is a schematic diagram of a gesture detection system of the present application;
FIG. 2 is a schematic process diagram of the gesture detection method of the present application;
FIG. 3 is a schematic diagram illustrating a parallel acceleration mode in the gesture detection method according to the present application;
FIG. 4 is a schematic illustration of the posture improvement procedure of the present application;
FIG. 5 is a schematic view of the posture correction mode of the present application;
in the figure, 1-an image acquisition unit, 2-an image detection unit, 3-a result output unit, 4-a half-body posture detection module, 5-an image correction module, 6-a whole body posture detection module, 7-a rotation sub-module and 8-an image correction judgment sub-module.
Detailed Description
Hereinafter, specific embodiments of the present application will be described in detail with reference to the accompanying drawings, and it will be apparent to those skilled in the art from this detailed description that the present application can be practiced. Features from different embodiments may be combined to yield new embodiments, or certain features may be substituted for certain embodiments to yield yet further preferred embodiments, without departing from the principles of the present application.
An API is a calling interface that an operating system leaves for an application program, which causes the operating system to execute commands of the application program by calling the API of the operating system.
Tensorflow is a second generation artificial intelligence learning system developed by Google based on DistBerief, and the naming of the Tensorflow comes from the operation principle of the Tensorflow. Tensor means an N-dimensional array, Flow means computation based on a dataflow graph, and TensorFlow is a computation process in which tensors Flow from one end of the Flow graph to the other. TensorFlow is a system that transports complex data structures into artificial intelligent neural networks for analysis and processing.
The TensorFlow can be used in the field of deep learning of multiple machines such as voice recognition or image recognition, various improvements are made on a deep learning infrastructure DistBeief developed in 2011, and the TensorFlow can be operated on various devices such as a smart phone and thousands of data center servers. TensorFlow will be completely open source and available to anyone.
Referring to fig. 1 to 5, the present application provides an attitude detection system, which includes an image acquisition unit, an image detection unit, and a result output unit, which are connected in sequence, wherein the image detection unit includes a half-length attitude detection module, an image correction module, and a whole-body attitude detection module, which are connected in sequence, the half-length attitude detection module is connected to the image acquisition unit, the half-length attitude detection module is connected to the whole-body attitude detection module, and the whole-body attitude detection module is connected to the result output unit.
Further, the half-body posture detection module is a lightweight posture detection module, and the half-body posture detection module is used for detecting the head, the two shoulders and the two crotches.
Further, the body posture detection module comprises a plurality of layers of neural networks.
Further, the image rectification module includes a rotation sub-module.
Further, the bust posture detection module includes an image rectification determination sub-module.
The application also provides a posture detection method, which comprises the following steps:
step 1: acquiring an image, and detecting the half-body posture of the image;
step 2: determining whether the image requires rectification;
and step 3: detecting the posture of the whole body of the image which does not need to be corrected and outputting a detection result; and (4) carrying out image correction on the image to be corrected, carrying out whole body posture detection, and outputting a detection result.
Human body reference line: in a large number of practices, it has been found that a pair of parallel lines and two perpendicular bisectors defined by the head and the shoulders and the crotch are almost unchanged in this process, although most human postures are different due to different motion states of the human body.
As shown in fig. 5, left 2 is a diagram of the gesture detection effect of a picture, and by the above method, we have found out the features of the head, two shoulders, two thighs, etc. In fig. 4, left 3, the present application indicates a pair of parallel lines (P1, P2) and two perpendicular bisectors (V1, P2) (V1, P1) by three dotted lines P1, P2 and V1 respectively, that is, P1 is parallel to P2, V1 is perpendicular to P1, and V1 is perpendicular to P2.
And (3) correction rules:
according to the right 1 display in fig. 4, the parallel lines are corrected to the angle perpendicular to the bottom of the current picture, i.e. the angle of the whole picture is rotated so that the portrait is finally perpendicular to the picture.
After determining the left 3 of fig. 4, the clockwise rotation is performed by the acute angle θ in the manner of fig. 5 until the dashed V1 line is perpendicular to the screen, so that the two dashed lines P1 and P2 are parallel to the screen.
In experiments and practice, if the pedestrian posture is perpendicular to the picture, the detection result will be more accurate, and in order to achieve the purpose, a simple half-body posture detector, also called a posture detection light-weight network, is firstly trained to detect only five points of the head, two shoulders and two glutes of the upper half-body. Because the network needs only a few simple layers of neural networks to be implemented. If the detection result shows that the person is not vertical to the picture at the moment, a correction algorithm is started to correct the posture, and the whole body detection is carried out after the correction is finished, so that the detection accuracy is improved. The overall detection process is shown in fig. 2.
Further, in the image in the step 2, if the portrait is perpendicular to the image, no correction is needed, and if the portrait is not perpendicular to the image, a correction is needed.
Further, the image is rectified into a rotation image in the step 3, so that the portrait is vertical to the image.
Further, the image rectification is realized through an API interface.
Further, the attitude detection method is based on a neural network basic computing platform, and the neural network computing platform is deployed on a TensorFlow open source framework with a distributed function.
The half-body posture detector first detects five points, namely the head, the two shoulders and the two crotch, and because the five points usually have obvious targets, a light-weight posture detection network can be completed. After detection, the computer system executes two tasks in parallel: if the image needs to be corrected, the image is sent into an image correction branch for correction, and after the correction is finished, the posture of the whole body is recognized; and if the posture correction is not needed, the posture detection is directly carried out. And finally outputting the detection result of the posture of the whole body.
In order to make the set of detection system more practical, the task scheduling mode of the invention on the general-purpose computing platform is shown in fig. 3. A conventional neural network computing platform is deployed on a TensorFlow open source framework with a distributed function, and various computing devices such as GPUs are manually loaded to be matched with computing tasks.
Two sets of algorithms with different resource occupancy rates are deployed on a general computing platform in parallel, resource allocation is written in a dynamic mode, and conventionally, a whole body posture detection network usually needs to consume more resources, so the proportion is high. The results of the next part of images (the part of images which do not need to be corrected) are detected in parallel and directly output to the right gray result part; and the other part detects that the image needs to be corrected, and at the moment, the CPU resource (an image correction API interface) is called to carry out rotation correction on the image. And sending the rotated image into a whole body detection network of the general computing platform again for detection, and outputting a result after the detection is finished.
The maximum advantage of the computing mode is that each module is guaranteed to work instead of waiting serially at any moment, computing time is saved to the maximum extent while computing resources are guaranteed to run at full load, and efficient processing of videos is completed.
The application provides a correction method for attitude detection, which improves the accuracy of attitude detection. Meanwhile, in comparison with similar algorithms, the method for effectively and quickly detecting the attitude based on the general computing platform is provided. The method is used together with a computing frame, is beneficial to realizing better detection effect on more devices by a network, and can further reduce the actual use cost.
According to the method, the precision is improved mainly in the attitude detection method, and meanwhile, the network speed is optimized in a targeted manner by combining the existing platform, so that the network is more accurate and faster.
In order to correct the problem of accuracy reduction caused by abnormal postures, a correction mechanism is introduced to improve the accuracy; the method and the device realize the full process of posture detection by a correction detection mode and design a corresponding computing system in a matching way, so that the method and the device can run faster on a general computing platform.
Although the present application has been described above with reference to specific embodiments, those skilled in the art will recognize that many changes may be made in the configuration and details of the present application within the principles and scope of the present application. The scope of protection of the application is determined by the appended claims, and all changes that come within the meaning and range of equivalency of the technical features are intended to be embraced therein.
Claims (10)
1. An attitude sensing system characterized by: the image detection unit comprises a half-body posture detection module, an image correction module and a whole-body posture detection module which are connected in sequence, the half-body posture detection module is connected with the image acquisition unit, the half-body posture detection module is connected with the whole-body posture detection module, and the whole-body posture detection module is connected with the result output unit.
2. The attitude detection system according to claim 1, characterized in that: the half-body posture detection module is a light-weight posture detection module and is used for detecting the head, the two shoulders and the two crotches.
3. The attitude detection system according to claim 2, characterized in that: the half-body posture detection module comprises a plurality of layers of neural networks.
4. The attitude detection system according to claim 1, characterized in that: the image rectification module includes a rotation sub-module.
5. The attitude detection system according to any one of claims 1 to 4, characterized in that: the half-body posture detection module comprises an image correction judgment submodule.
6. An attitude detection method, characterized by: the method comprises the following steps:
step 1: acquiring an image, and detecting the half-body posture of the image;
step 2: determining whether the image requires rectification;
and step 3: detecting the posture of the whole body of the image which does not need to be corrected and outputting a detection result; and (4) carrying out image correction on the image to be corrected, carrying out whole body posture detection, and outputting a detection result.
7. The attitude detection method according to claim 6, characterized in that: in the step 2, if the portrait is perpendicular to the image, no correction is needed, and if the portrait is not perpendicular to the image, the correction is needed.
8. The attitude detection method according to claim 6, characterized in that: and in the step 3, the image is rectified into a rotation image so that the portrait is vertical to the image.
9. The attitude detection method according to claim 7, characterized in that: the image rectification is realized through an API interface.
10. The attitude detection method according to any one of claims 6 to 9, characterized by: the attitude detection method is based on a neural network basic computing platform, and the neural network computing platform is deployed on a TensorFlow open source framework with a distributed function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911232459.3A CN111144229A (en) | 2019-12-05 | 2019-12-05 | Posture detection system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911232459.3A CN111144229A (en) | 2019-12-05 | 2019-12-05 | Posture detection system and method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111144229A true CN111144229A (en) | 2020-05-12 |
Family
ID=70517660
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911232459.3A Pending CN111144229A (en) | 2019-12-05 | 2019-12-05 | Posture detection system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111144229A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107886074A (en) * | 2017-11-13 | 2018-04-06 | 苏州科达科技股份有限公司 | A kind of method for detecting human face and face detection system |
| JP2018073385A (en) * | 2016-10-22 | 2018-05-10 | 俊之 坂本 | Image processing device and program |
| CN109919141A (en) * | 2019-04-09 | 2019-06-21 | 广东省智能制造研究所 | A kind of recognition methods again of the pedestrian based on skeleton pose |
| CN110378871A (en) * | 2019-06-06 | 2019-10-25 | 绍兴聚量数据技术有限公司 | Game charater original painting copy detection method based on posture feature |
-
2019
- 2019-12-05 CN CN201911232459.3A patent/CN111144229A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2018073385A (en) * | 2016-10-22 | 2018-05-10 | 俊之 坂本 | Image processing device and program |
| CN107886074A (en) * | 2017-11-13 | 2018-04-06 | 苏州科达科技股份有限公司 | A kind of method for detecting human face and face detection system |
| CN109919141A (en) * | 2019-04-09 | 2019-06-21 | 广东省智能制造研究所 | A kind of recognition methods again of the pedestrian based on skeleton pose |
| CN110378871A (en) * | 2019-06-06 | 2019-10-25 | 绍兴聚量数据技术有限公司 | Game charater original painting copy detection method based on posture feature |
Non-Patent Citations (1)
| Title |
|---|
| 雷洪编著: "《多核异构并行计算OpenMP4.5 C_C++篇》", 30 April 2018, 冶金工业出版社, pages: 14 - 15 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11741736B2 (en) | Determining associations between objects and persons using machine learning models | |
| CN109740534B (en) | Image processing method, device and processing equipment | |
| CN109791625B (en) | Facial recognition using artificial neural networks | |
| CN108875522B (en) | Face clustering method, device and system and storage medium | |
| CN111062239A (en) | Human body target detection method and device, computer equipment and storage medium | |
| WO2020015752A1 (en) | Object attribute identification method, apparatus and system, and computing device | |
| WO2017000816A1 (en) | Face recognition system and face recognition method | |
| CN112131985B (en) | Real-time light human body posture estimation method based on OpenPose improvement | |
| US20220262093A1 (en) | Object detection method and system, and non-transitory computer-readable medium | |
| CN111654694A (en) | Quality evaluation method and device of image processing algorithm and electronic equipment | |
| Damdoo et al. | Adaptive hand gesture recognition system using machine learning approach | |
| CN114332927A (en) | Classroom hand-raising behavior detection method, system, computer equipment and storage medium | |
| CN112307872B (en) | Method and device for detecting target object | |
| CN116580169B (en) | Digital man driving method and device, electronic equipment and storage medium | |
| Shi et al. | An FPGA-based smart camera for gesture recognition in HCI applications | |
| CN116453222B (en) | Target object posture determining method, training device and storage medium | |
| CN111339920A (en) | Cash adding behavior detection method, device and system, storage medium and electronic terminal | |
| Shi et al. | Fbi-pose: Towards bridging the gap between 2d images and 3d human poses using forward-or-backward information | |
| CN111144229A (en) | Posture detection system and method | |
| CN116246343A (en) | Light human body behavior recognition method and device | |
| Yamashita et al. | Facial point detection using convolutional neural network transferred from a heterogeneous task | |
| WO2020207294A1 (en) | Service processing method and apparatus, and storage medium and electronic device | |
| CN105323920B (en) | A kind of Lighting Control Assembly and method based on human body and its position detection | |
| Xue et al. | MPE-DETR: A multiscale pyramid enhancement network for object detection in low-light images | |
| CN113255408B (en) | Behavior recognition method, behavior recognition device, electronic equipment and storage medium |
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 |