CN115909457A - Mask wearing detection method based on polarization imaging AI recognition - Google Patents
Mask wearing detection method based on polarization imaging AI recognition Download PDFInfo
- Publication number
- CN115909457A CN115909457A CN202211471689.7A CN202211471689A CN115909457A CN 115909457 A CN115909457 A CN 115909457A CN 202211471689 A CN202211471689 A CN 202211471689A CN 115909457 A CN115909457 A CN 115909457A
- Authority
- CN
- China
- Prior art keywords
- polarization
- images
- polarized
- imaging
- detection method
- 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
- 230000010287 polarization Effects 0.000 title claims abstract description 78
- 238000001514 detection method Methods 0.000 title claims abstract description 32
- 238000003384 imaging method Methods 0.000 title claims abstract description 27
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 12
- 238000013135 deep learning Methods 0.000 claims abstract description 10
- 230000006870 function Effects 0.000 claims abstract description 6
- 239000000463 material Substances 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 10
- 230000000694 effects Effects 0.000 claims description 8
- 238000002372 labelling Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 230000014509 gene expression Effects 0.000 claims description 4
- 230000003287 optical effect Effects 0.000 claims description 3
- 238000000034 method Methods 0.000 abstract description 20
- 238000010586 diagram Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 239000004744 fabric Substances 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 230000002265 prevention Effects 0.000 description 2
- 238000005728 strengthening Methods 0.000 description 2
- 230000003313 weakening effect Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000001815 facial effect Effects 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 239000002657 fibrous material Substances 0.000 description 1
- 239000010408 film Substances 0.000 description 1
- 239000004750 melt-blown nonwoven Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000002120 nanofilm Substances 0.000 description 1
- 239000004745 nonwoven fabric Substances 0.000 description 1
- 230000036544 posture Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000011895 specific detection Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a method for identifying mask wearing based on polarization imaging AI, which comprises the following steps: acquiring polarization images of a target to be detected from different angles and different environments in real time; analyzing the characteristics of the polarization images, and classifying the polarization images of various types; carrying out deep learning on the labeled and classified polarization images by using a CNN (convolutional neural network) so that an algorithm model can distinguish whether a mask is worn or not; and adding a cloud early warning function based on the trained and perfected model. Based on the vividness of the imaging characteristics of the polarization camera, the invention improves the identification precision by combining the polarized light imaging with the AI intelligent algorithm, avoids the problems of easy overexposure, excessive darkness, easy interference and the like in the image identification of the traditional RGB area-array camera, optimizes the detection method and has better timeliness realized on the algorithm.
Description
Technical Field
The invention relates to the field of optics, in particular to a method and a system for recognizing mask wearing detection based on polarization imaging AI.
Background
At present, applications and equipment for detecting the mask for wearing are published in the market, some relatively mature detection methods such as the well-known Viola-Jones algorithm, the deep learning algorithm SSD, YOLO and the like are provided, and most detection modes are that a three-primary-color RGB camera is used for identifying by rapidly scanning crowds and matching with an algorithm of an artificial intelligence system.
The traditional camera can be used for detecting whether the mask is worn or not, but has limitations. The traditional RGB image recognition method has the following problems before the perfect training is not achieved:
(1) The contrast of the dataset features of the ordinary intensity image is not high. The characteristics are not outstanding, the nose and the mouth are easily shielded by objects, and the model is considered as the problem of wearing the mask;
(2) The RGB camera is obviously influenced by light environment. During identification, strong light irradiation overexposure or weak light irradiation too dark may occur, which may cause difficulty in information acquisition and may not be identified.
(3) Some masks with patterns may cause misjudgment. For example, when there are other faces in the mask pattern or the mask pattern is just the lower half of a face, the recognition algorithm is prone to cause erroneous judgment.
In order to reduce interference, technicians also need to optimize an AI algorithm and introduce more sample pictures to improve the sample evaluation accuracy, background technicians have large workload and need rapid detection equipment, and therefore, the method disclosed by the invention fundamentally utilizes a polarized light imaging technology to identify and detect whether the mask is worn or not.
Disclosure of Invention
The invention mainly solves the technical problem of providing a mask wearing detection method based on polarization imaging AI identification, which comprises the following steps:
s1, acquiring polarization images of a target to be detected in real time from different angles and different environments by using a polarization camera;
and S2, analyzing the characteristics of the polarization images, classifying the polarization images of various types, and marking the expressions of the masks with different patterns, materials and shapes on the face when the collected polarization image samples reach a certain number.
And step S3: carrying out deep learning on the labeled and classified polarization images by using a CNN convolutional neural network, so that an algorithm model can distinguish whether the mask is worn or not;
and S4, evaluating the comprehensive performance of the model according to the mAP value of the model and the actual recognition prediction result, and repeating training to improve the recognition effect.
And S5, adding a cloud early warning function based on the trained and perfected algorithm model.
Preferably, the polarized image is obtained by using a set of polarization imaging device built by an optical wide-angle zoom lens and a polarization camera to obtain a real-time dynamic high-precision polarized image.
Preferably, the polarization imaging device integrates a polarization element with four direction angles on the surface of the photosensitive element, and four images with the same resolution and different polarization states of the same object can be obtained by one-time shooting.
Preferably, the polarization imaging device collects images of four-way polarization angles of 0 °, 45 °, 90 °, and 135 ° of the target to be detected.
Preferably, the deep learning is to integrate images of different categories for two-classification labeling, and introduce a YOLOv5 algorithm model to perform feature extraction on the polarized image, so as to finally realize target detection.
The invention has the beneficial effects that:
(1) In some specific occasions, such as poor illumination condition, or difficult direct detection of detected objects by using a traditional area array camera, the polarization information which cannot be detected by a normal image sensor can be captured by using the polarization of light. The effects of strong light weakening and weak light strengthening are realized, so that the polarization camera can be used for avoiding the influence of too strong light or too weak light on imaging, and the image features can be distinct.
(2) The epidemic prevention detection system under the polarization camera has different sensitivities to the polarized light due to different materials, and shows different performances when the polarization camera images, so that the characteristics can be utilized to distinguish articles made of different materials, and the problem of misjudgment of a plurality of AI based on the RGB camera can be well solved.
(3) Aiming at the algorithm for identifying and detecting the mask information under polarization imaging, the detection method is optimized, and the timeliness realized on the algorithm is better.
Drawings
Fig. 1 is a block diagram of a mask wearing system based on polarization imaging AI identification according to an embodiment of the present invention;
FIG. 2 is a polarized image contrast diagram of a typical worn mask and a typical unworn mask;
fig. 3 is an identification effect diagram of a mask wearing detection method based on polarization imaging AI identification in an embodiment of the present invention.
Detailed Description
The invention mainly solves the technical problem of providing a mask wearing detection method based on polarization imaging AI recognition, and introduces a data classification set into a trained convolutional neural network model so that the model can accurately carry out reasoning prediction on a detected target. The implementation process comprises the following steps:
s1, acquiring polarization images of a target to be detected in real time from different angles and different environments by using a polarization camera;
s2, distinguishing the acquired images, analyzing the characteristics of the polarization images, and classifying the polarization images of various types;
s3, when a large number of polarized image samples are collected, labeling and classifying the expressions of masks with different patterns, materials and shapes on the face, and performing deep learning on the labeled and classified polarized images by using a CNN convolutional neural network so that an algorithm model can distinguish whether the masks are worn or not;
and S4, evaluating the comprehensive performance of the model according to the mAP value of the model and the actual recognition prediction result, and repeating training to improve the recognition effect.
And S5, adding a cloud early warning function based on the trained and perfected algorithm model.
It should be noted that: the polarized image acquisition device used by the invention is an SONY polarized image sensor, has high resolution, can acquire four-angle target polarized images in real time, can analyze four-angle gray level images, and has high integration level and compact volume.
When a polarization camera is used for acquiring image data, the degree of polarization is the most fundamental information and is an important parameter for measuring polarization information in electromagnetic waves, and we customarily describe the degree of polarization as the proportion of polarized light in total light intensity. And the Stokes vector is a mathematical principle for calculating the degree of polarization and is a mathematical representation of the state of optical polarization. The method mainly comprises four vector parameters as parameters of polarization information:
since the circular polarization component is few and almost negligible in a practical environment, V =0 can be obtained in the calculation process of the stokes vector, and the degree of polarization of the polarization image can be calculated by using formula (2).
The four-direction polarization state image which can be obtained by the sensor is an image of a target to be detected at four different polarization angles of 0 degree, 45 degrees, 90 degrees and 135 degrees, and the four polarization images are respectively recorded as I 0 、I 45 、I 90 、I 135 . The polarization is characterized by means of the stokes vector.
In the present invention, the polarization image is not limited to the four polarization images of 0 °, 45 °, 90 ° and 135 ° described above, and may be a different polarization image of five or more angles.
According to present common masks mainly divided into melt-blown non-woven fabric masks and nano film masks, the material components mainly comprise melt-blown fabrics, non-woven fabrics or film fiber materials with the diameter of 100-200 nanometers, because the molecular arrangement of mask fibers usually has different orientations, and has birefringence and dichroism to polarized light, when gray level images of a human face and the masks can be observed by utilizing a polarization camera, the contrast is very high, and the face, the masks and the masks can be identified favorably.
Specifically, because gauze mask and people's face show the difference in intensity value and intensity distribution in the polarization image, through modes such as degree of depth study and feature extraction, to the gauze mask characteristic of different patterns, different properties, different materials and the feature integration extraction of people's face to distinguish the difference of the two in continuous model training, thereby realize the discernment whether to wear the gauze mask.
Please refer to fig. 1, which illustrates an embodiment of the present invention including:
a mask wearing detection method based on polarization imaging AI identification is used for a system for epidemic prevention mask detection, aiming at the problems that a polarization camera is bright in imaging and less in interference, can distinguish the characteristics of materials, effectively avoids the problems in image identification of a traditional RGB area-array camera and the like, and obtains a to-be-detected photo.
The specific detection steps are as follows:
s1: the method comprises the steps of utilizing a polarization camera to collect different shooting environments and different shooting angles, classifying the expressions of masks with different patterns, materials and shapes on a human face, obtaining about 2000 polarization images, and integrating and labeling the images.
S2: and classifying and identifying by using a CNN type YOLOv5 algorithm, and finally distinguishing the mask from the face. The prediction of the YOLO algorithm is single-stage, namely, when an image is input, the result can be output only through a CNN (convolutional neural network), and compared with two-stage algorithms such as DPM (differential pulse width modulation) and fast-Rcnn, the work of a preselected frame is omitted, so that the prediction speed of the YOLO algorithm is higher, and the method is suitable for target detection of real-time tracking.
Wherein, the specific training process comprises a prediction phase and a training phase.
In the prediction stage, a polarization image is input on the basis of training, the polarization image is divided into grids with different sizes, each grid comprises a plurality of priori frames, and the convolution layer down-sampling and feature extraction are carried out on the image information through a backbone structure named backbone. And classifying through a full connecting layer of a tack structure, wherein the extracted and classified characteristic layer is used for classification prediction and regression prediction.
In the training stage, the process is a process of continuously updating reversely, iterating network parameters, gradually reducing the gradient of the loss value, further finely adjusting the weight in the neural network, and minimizing the loss function. The ultimate goal is to fit or approximate the prior box to the true box continuously. The method is a continuous regression process, a continuous value can be predicted in the process and then is compared with a labeled value, and the closer the predicted value and the labeled value are, the better the effect is. The specific regression parameters are:
(1) Position error:
(2) Confidence (confidence) error:
(3) Object class error:
s3: and after the polarized images are labeled and classified, importing the classified polarized images into a YOLOv5 model, and further training according to mAP (maximum image probability) values and TensorBoard feedback information provided by the model after the polarized images can basically distinguish image information of expected polarized wearing masks and non-wearing masks so as to improve the identification precision of the YOLOv5 algorithm model.
S4: the model which is trained well is deployed in an actual scene for application, the actual prediction classification effect is observed, as shown in fig. 3, the model can meet the expected mask identification, and meanwhile, the real-time accurate identification under the conditions of multiple scenes and multiple persons can be realized.
S5: after the model is trained, the model can be sent to a cloud for real-time alarm processing.
The method uses a YOLOv5s deep learning algorithm in combination with polarization imaging characteristics, has strong robustness, and can realize detection tasks aiming at different illumination environments and different sampling angles. Meanwhile, the weight file of YOLOv5 is 27 megabytes, the model is light in size, and compared with other identification methods, the model is easier to transplant and embed.
The invention is based on a single-stage detection method, and can realize rapid and stable real-time mask detection and tracking. The adopted DIOU _ nms structure can effectively eliminate the missing detection of the overlapped target and the multi-sample target. Meanwhile, in order to ensure the accuracy, the deep learning method adopts the weighting of IOU, GIOU, DIOU and CIOU.
The invention adopts a method of sampling by a polarization camera, obtains polarization information of different forms and different materials, obtains polarization image information aiming at various mask wearing modes, various head postures and various facial shelters or accessories creatively, and classifies and trains the information to ensure that the confidence coefficient of a final prediction result tends to 100%.
The invention can finally realize the real-time multi-person living body detection of more than 30fps, the detection distance can reach 5 meters, the effects of strong light weakening and weak light strengthening can be realized, the influence of too strong light or too weak light on imaging can be avoided, and the image can be clear and has distinct characteristics.
Therefore, the invention can be applied to people stream detection and checkpoint detection, such as railway stations, airports, hospitals and other environments. The automatic, efficient, accurate and intelligent face living body detection is realized, and the safety protection of the biological recognition system is improved.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes performed by the present specification and drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (5)
1. A mask wearing detection method based on polarization imaging AI identification is characterized by comprising the following steps:
s1, collecting polarization images, and acquiring the polarization images of a target to be detected in real time from different angles and different environments by using a polarization camera;
s2, image labeling and classifying, namely analyzing the characteristics of the polarization images, classifying the polarization images of various types, and labeling the expressions of masks with different patterns, materials and shapes on the face when the collected polarization image samples reach a certain number;
s3, deep learning of images, namely performing deep learning on the labeled and classified polarization images by using a CNN (convolutional neural network) so that an algorithm model can distinguish whether a mask is worn or not;
s4, improving the performance of the model, evaluating the comprehensive performance of the model according to the mAP value of the algorithm model and the actual recognition prediction result, and repeatedly training to improve the recognition effect;
and S5, adding an additional function, and adding a cloud early warning function based on the trained and perfected algorithm model.
2. The AI identification mask wearing detection method based on polarized imaging according to claim 1, wherein the polarized image is obtained by a set of polarized imaging device constructed by an optical wide-angle zoom lens and a polarized camera, so as to obtain a real-time dynamic high-precision polarized image.
3. The AI identification mask wearing detection method according to claim 2, wherein the polarized imaging device is formed by integrating a polarized element with four directions and angles on the surface of a photosensitive element, so that four images with the same resolution and different polarization states of the same object can be obtained by one-time shooting.
4. The AI identification mask wearing detection method based on polarization imaging as claimed in claim 2, wherein the polarization imaging device collects images of 0 °, 45 °, 90 °, 135 ° four-way polarization angles of the target to be detected.
5. The AI identification mask wearing detection method based on polarization imaging as claimed in claim 1, wherein: the deep learning is to integrate images of different categories for two-classification labeling, and introduce a YOLOv5 algorithm model to perform feature extraction on the polarized images, so as to finally realize target detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211471689.7A CN115909457A (en) | 2022-11-23 | 2022-11-23 | Mask wearing detection method based on polarization imaging AI recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211471689.7A CN115909457A (en) | 2022-11-23 | 2022-11-23 | Mask wearing detection method based on polarization imaging AI recognition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115909457A true CN115909457A (en) | 2023-04-04 |
Family
ID=86474151
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211471689.7A Pending CN115909457A (en) | 2022-11-23 | 2022-11-23 | Mask wearing detection method based on polarization imaging AI recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115909457A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111524080A (en) * | 2020-04-22 | 2020-08-11 | 杭州夭灵夭智能科技有限公司 | Face skin feature identification method, terminal and computer equipment |
| CN111539348A (en) * | 2020-04-27 | 2020-08-14 | 天津中科智能识别产业技术研究院有限公司 | Face living body detection method based on polarization imaging |
| CN113610050A (en) * | 2021-08-26 | 2021-11-05 | 齐鲁工业大学 | Mask wearing real-time detection method based on YOLOv5 |
| CN114399799A (en) * | 2021-11-22 | 2022-04-26 | 电子科技大学 | Mask wearing detection method based on YOLOv5 network |
-
2022
- 2022-11-23 CN CN202211471689.7A patent/CN115909457A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111524080A (en) * | 2020-04-22 | 2020-08-11 | 杭州夭灵夭智能科技有限公司 | Face skin feature identification method, terminal and computer equipment |
| CN111539348A (en) * | 2020-04-27 | 2020-08-14 | 天津中科智能识别产业技术研究院有限公司 | Face living body detection method based on polarization imaging |
| CN113610050A (en) * | 2021-08-26 | 2021-11-05 | 齐鲁工业大学 | Mask wearing real-time detection method based on YOLOv5 |
| CN114399799A (en) * | 2021-11-22 | 2022-04-26 | 电子科技大学 | Mask wearing detection method based on YOLOv5 network |
Non-Patent Citations (2)
| Title |
|---|
| JOSEPH REDMONET AL.: "You Only Look Once:Unified, Real-Time Object Detection", 《ARXIV》, pages 1 - 10 * |
| 邵洪禹等: "地基大口径超高分辨偏振望远光学系统设计", 《中国优秀硕士学位论文全文数据库 基础科学辑》, no. 01, pages 2 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109819208B (en) | Intensive population security monitoring management method based on artificial intelligence dynamic monitoring | |
| Gajjar et al. | Human detection and tracking for video surveillance: A cognitive science approach | |
| US11263435B2 (en) | Method for recognizing face from monitoring video data | |
| CN107123131B (en) | Moving target detection method based on deep learning | |
| KR100974293B1 (en) | METHOD AND SYSTEM FOR AUTOMATED FACE DETECTION and Recognition | |
| CN106778609A (en) | A kind of electric power construction field personnel uniform wears recognition methods | |
| CN110728252B (en) | Face detection method applied to regional personnel motion trail monitoring | |
| CN108960047A (en) | Face De-weight method in video monitoring based on the secondary tree of depth | |
| CN109522853A (en) | Face datection and searching method towards monitor video | |
| CN106951889A (en) | Underground high risk zone moving target monitoring and management system | |
| CN103310200B (en) | Face identification method | |
| CN103034838A (en) | Special vehicle instrument type identification and calibration method based on image characteristics | |
| CN102622614B (en) | Knife switch closing reliability judging method based on distance between knife switch arm feature point and fixing end | |
| WO2021217764A1 (en) | Human face liveness detection method based on polarization imaging | |
| CN113269091A (en) | Personnel trajectory analysis method, equipment and medium for intelligent park | |
| Krišto et al. | Thermal imaging dataset for person detection | |
| CN106548131A (en) | A kind of workmen's safety helmet real-time detection method based on pedestrian detection | |
| CN109271932A (en) | Pedestrian based on color-match recognition methods again | |
| CN110188715A (en) | A kind of video human face biopsy method of multi frame detection ballot | |
| CN112541403B (en) | Indoor personnel falling detection method by utilizing infrared camera | |
| CN112037252A (en) | Eagle eye vision-based target tracking method and system | |
| CN114898238A (en) | Wild animal remote sensing identification method and device | |
| CN115620090A (en) | Model training method, low-illumination target re-recognition method and device and terminal equipment | |
| Lopar et al. | An overview and evaluation of various face and eyes detection algorithms for driver fatigue monitoring systems | |
| CN108073873A (en) | Human face detection and tracing system based on high-definition intelligent video camera |
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 |