CN113378767A - Elevator early warning control method and system based on adaptive learning - Google Patents
Elevator early warning control method and system based on adaptive learning Download PDFInfo
- Publication number
- CN113378767A CN113378767A CN202110716814.5A CN202110716814A CN113378767A CN 113378767 A CN113378767 A CN 113378767A CN 202110716814 A CN202110716814 A CN 202110716814A CN 113378767 A CN113378767 A CN 113378767A
- Authority
- CN
- China
- Prior art keywords
- target detection
- detection model
- fine
- tuning
- image
- 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
- 238000000034 method Methods 0.000 title claims abstract description 36
- 230000003044 adaptive effect Effects 0.000 title claims description 12
- 238000001514 detection method Methods 0.000 claims abstract description 177
- 238000012216 screening Methods 0.000 claims abstract description 55
- 238000012549 training Methods 0.000 claims description 78
- 238000012360 testing method Methods 0.000 claims description 25
- 238000012795 verification Methods 0.000 claims description 11
- 238000013101 initial test Methods 0.000 claims description 6
- 238000009825 accumulation Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 241000282472 Canis lupus familiaris Species 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000011895 specific detection Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B66—HOISTING; LIFTING; HAULING
- B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
- B66B5/00—Applications of checking, fault-correcting, or safety devices in elevators
- B66B5/0006—Monitoring devices or performance analysers
- B66B5/0018—Devices monitoring the operating condition of the elevator system
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B66—HOISTING; LIFTING; HAULING
- B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
- B66B5/00—Applications of checking, fault-correcting, or safety devices in elevators
- B66B5/0006—Monitoring devices or performance analysers
- B66B5/0037—Performance analysers
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B66—HOISTING; LIFTING; HAULING
- B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
- B66B5/00—Applications of checking, fault-correcting, or safety devices in elevators
- B66B5/0087—Devices facilitating maintenance, repair or inspection tasks
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B66—HOISTING; LIFTING; HAULING
- B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
- B66B5/00—Applications of checking, fault-correcting, or safety devices in elevators
- B66B5/02—Applications of checking, fault-correcting, or safety devices in elevators responsive to abnormal operating conditions
-
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Indicating And Signalling Devices For Elevators (AREA)
Abstract
The invention relates to an elevator early warning control method and system based on self-adaptive learning, wherein the elevator early warning control method comprises the following steps: s1, collecting video image information of an elevator, and performing online target detection on the video image information by adopting a target detection model to obtain a detection result; wherein the detection result comprises: a first image in which an object is detected and a second image in which an object is not detected; s2, controlling and giving an alarm to the elevator based on the detection result; s3, uploading all or part of the first image and all or part of the second image for screening by a user and acquiring a screening result; wherein the screening result is a false detection or missing detection image in the first image and the second image; and S4, updating the target detection model by using the screening result. The invention has high detection precision and good early warning effect.
Description
Technical Field
The invention relates to an elevator early warning control method and an elevator early warning control system, in particular to an elevator early warning control method and an elevator early warning control system based on self-adaptive learning.
Background
In recent years, higher-level cells have been increasing. In order to prevent fire-fighting hidden dangers, the provision of forbidding a battery car to enter an elevator is provided in most high-rise communities at present; in addition, the individual high-grade district can also prohibit bicycles or large dogs from taking the elevator. But all have the problem of great difficulty in implementation. Some methods of detection based on video data have emerged to prevent a particular target from riding an elevator.
However, although the method for preventing the elevator from being taken by the elevator based on the video image detection can effectively prevent the specific class of objects from entering the elevator, the method also has some problems:
1) due to the accuracy problem, most schemes only realize the alarm, and if the elevator taking people are blind to alarm, the elevator can still be taken normally;
2) even if an elevator control scene exists, targets easy to detect cannot be taken after being stopped for taking the elevator for multiple times, and finally, targets easy to detect mistakenly and miss detection are remained, so that the accuracy in the actual scene is lower and lower;
3) the scheme of collecting difficult samples by partial manual work has long time, large difficulty and higher labor cost.
Disclosure of Invention
The invention aims to provide an elevator early warning control method and system based on adaptive learning, and solves the problem of inaccurate detection.
In order to achieve the aim, the invention provides an elevator early warning control method based on adaptive learning, which comprises the following steps of:
s1, collecting video image information of an elevator, and performing online target detection on the video image information by adopting a target detection model to obtain a detection result; wherein the detection result comprises: a first image in which an object is detected and a second image in which an object is not detected;
s2, controlling and giving an alarm to the elevator based on the detection result;
s3, uploading all or part of the first image and all or part of the second image for screening by a user and acquiring a screening result; wherein the screening result is a false detection or missing detection image in the first image and the second image;
and S4, updating the target detection model by using the screening result.
According to an aspect of the present invention, in step S1, the step of acquiring video image information of an elevator, and performing online target detection on the video image information by using a target detection model, and acquiring a detection result includes:
s11, acquiring the first image with the target and the second image without the target in the video image information based on a first threshold value;
and S12, acquiring an uncertain target image in the first image based on a second threshold value.
According to an aspect of the present invention, the step of updating the target detection model with the screening result in step S4 includes:
s41, acquiring the screening result and the uncertain target image, and counting the data volume of the uncertain target image;
s42, if the data volume reaches a first preset value, selectively combining the screening result, the uncertain target image and an initial sample set for off-line training of the target detection model to obtain a fine-tuning training set for off-line fine tuning of the target detection model;
s43, performing off-line fine tuning training on the target detection model based on the fine tuning training set to obtain a fine tuning target detection model;
and S44, verifying the fine-tuning target detection model, and updating the target detection model running on line based on the parameters of the fine-tuning target detection model if the verification result meets the preset requirement.
According to an aspect of the present invention, in the step S44, in the step of verifying the fine-tuning target detection model, if the verification result does not satisfy the preset requirement, part of the previously acquired screening result and the uncertain target image are deleted, and the step S41 is executed again to continue acquiring new screening result and uncertain target image for data accumulation until the data amount reaches the first preset value again, and the steps S42 to S44 are executed.
According to an aspect of the present invention, in step S42, the step of selectively combining the screening result, the uncertain target image, and the initial sample set for offline training of the target detection model to obtain a fine-tuning training set for offline fine-tuning of the target detection model includes:
s421, rechecking the screening result and the data volume of the uncertain target image;
s422, extracting a certain amount of sample images in an initial training set of the initial sample set for off-line training the target detection model according to a preset proportion to serve as a first training set;
s423, dividing the screening result and the uncertain target image into a second training set and a first testing set according to the original training testing set proportion in the initial sample set;
and S424, combining the first training set and the second training set as the fine tuning training set, and combining the initial test set used for testing in the initial sample set and the first test set as the fine tuning test set.
According to an aspect of the present invention, in step S43, the step of performing offline fine-tuning training on the target detection model based on the fine-tuning training set to obtain a fine-tuning target detection model includes:
s431, performing offline fine tuning training on the target detection model by adopting an Adam optimizer; wherein the initial learning rate is set to 1e-5 and the minimum learning rate is set to 1 e-6;
and S432, training 5 rounds based on the fine tuning training set, wherein a Cosine learning rate adjusting method is used for adjusting the learning rate in the training process.
According to an aspect of the present invention, in step S44, the fine-tuning target detection model is verified, and if the verification result meets a preset requirement, the target detection model running on line is updated based on the parameters of the fine-tuning target detection model, where the preset requirement is that the average precision of the fine-tuning target detection model on the fine-tuning test set meets a preset threshold;
and if the average precision of the fine-tuning target detection model on the fine-tuning test set meets a preset threshold, replacing the parameters of the target detection model running on the line with the parameters of the fine-tuning target detection model.
According to an aspect of the present invention, s44, verifying the fine-tuning target detection model, and if the verification result meets a preset requirement, updating the target detection model running on line based on the parameters of the fine-tuning target detection model, further includes:
updating the initial sample set, comprising:
randomly extracting sample images with the same quantity as the first training set in the fine tuning training set, and putting the sample images into the initial sample set;
randomly sampling half of the sample images in the first test set to replace an equal number of images in the initial test set.
According to an aspect of the present invention, in step S1, the object detection model adopts yolov5 model, which outputs corresponding detection result for each inputted image;
the detection result is expressed as: n x (c, s, x)1,y1,x2,y2)；
Wherein N is the number of detected targets, c is the category of the detected targets, and s is the confidence of the detected targets; (x)1,y1,x2,y2) Coordinate positions of the upper left corner and the lower right corner of the rectangular frame in the image are detected.
In order to achieve the above object, the present invention provides a system using the aforementioned adaptive learning-based elevator early warning control method, including: the system comprises a data acquisition module, front-end equipment, an elevator controller, a user feedback module and a cloud server;
the data acquisition module is used for acquiring video image information of the elevator;
the front-end equipment transmits the video image information to the cloud server and receives a detection result sent by the cloud server;
the elevator controller is used for receiving the detection result and controlling and giving an alarm to the elevator;
the user feedback module is used for receiving the detection result uploaded by the front-end equipment and screening the detection result;
and the cloud server is used for receiving the screening result uploaded by the user feedback module, carrying out target detection on the video image information and outputting the detection result.
According to an aspect of the present invention, the cloud server includes: an incremental image data module and an adaptive learning module;
the incremental image data module is used for storing an initial sample set and the screening result for off-line training of the target detection model;
the self-adaptive learning module is used for performing offline fine-tuning training on the target detection model.
According to one scheme of the invention, the scheme combines huge data (namely false detection data and missed detection data) uploaded by users for fine adjustment and updating of the target detection model, so that the detection precision of the method becomes higher and higher along with the supplement of false detection and missed detection images, and the accurate early warning and control of the elevator are realized.
According to one scheme of the invention, in the service operation process of the server side, difficult samples can be continuously recorded; when the number of the difficult samples reaches a certain number; and integrating the difficult sample and part of the initial sample set data, and performing fine adjustment on the model parameters. Through constantly repeating this flow, the accuracy of high in the clouds detection service is further continuously promoted.
According to the scheme of the invention, the training set and the test set in the initial sample set are updated after the target detection model is finely adjusted, so that the continuous updating and accumulation of the sample set are ensured, the detection precision of the scheme is further increased along with the change of time, and the requirements of high efficiency and high precision are met.
According to one scheme of the invention, the screening of the user is combined, so that the scheme can be changed correspondingly according to the needs of the user, the requirements of the user can be further met, and the satisfaction degree of the user in the using process is improved.
Drawings
Fig. 1 schematically shows a block diagram of steps of an elevator warning control method according to an embodiment of the present invention;
fig. 2 schematically shows a target detection model update flow chart of an elevator early warning control method according to an embodiment of the present invention;
fig. 3 schematically shows a block diagram of an elevator early warning control system according to an embodiment of the present invention;
fig. 4 schematically shows a target detection flow diagram of an elevator early warning control system according to an embodiment of the present invention.
Detailed Description
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments will be briefly described below.
As shown in fig. 1, according to an embodiment of the present invention, an elevator early warning control method based on adaptive learning of the present invention includes the following steps:
s1, collecting video image information of an elevator, and performing online target detection on the video image information by adopting a target detection model to obtain a detection result; wherein, the detection result includes: a first image in which an object is detected and a second image in which an object is not detected;
s2, controlling and alarming the elevator based on the detection result;
s3, uploading all the first images and all or part of the second images for screening by the user and acquiring a screening result; the screening result is a false detection or missing detection image in the first image and the second image;
and S4, updating the target detection model by using the screening result.
According to one embodiment of the present invention, in step S1, video image information of the elevator is collected, and on-line target detection is performed on the video image information by using a target detection model, and in the step of acquiring the detection result, target detection is performed on each input image by using the target detection model. Wherein, the target detection model is a model generated by adopting yolov5 detection network, and outputs a corresponding detection result for each input image, and the detection result can be expressed as: n x (c, s, x)1,y1,x2,y2) (ii) a Wherein N is the number of detected targets, c is the category of the detected targets, and s is the confidence of the detected targets; (x)1,y1,x2,y2) Coordinate positions of the upper left corner and the lower right corner of the rectangular frame in the image are detected. The specific detection step of the target detection model comprises the following steps:
s11, acquiring a second image which is not detected by a first image with a target in the video image information based on a first threshold; in the present embodiment, only an image in which s is larger than a certain threshold value is output as the first image. In the present embodiment, s may be set to 0.3. The use of a smaller threshold allows the user to balance the detection rate and the precision rate by himself. For a certain elevator, if there are more false detections, a higher threshold may be used for further filtering.
And S12, acquiring an uncertain target image in the first image based on the second threshold value. In the present embodiment, the object detection model is applied to a part (for example, 10%) or all of the detected uncertain objects, that is, the images with confidence levels around a certain threshold (0.5 ± 0.05 in the present embodiment). In the present embodiment, the threshold value 0.5 set here is an empirical value for collecting indeterminate samples, and it is considered that the confidence of the positive samples determined should be close to 1 and the confidence of the negative samples determined should be close to 0.
According to one embodiment of the invention, in step S2, the elevator is controlled and alerted based on the detection result. In the present embodiment, when the detection result includes the first image of the detected target, a control signal is output to control the elevator. In this embodiment, the output control signal is a door opening and closing signal, and further, the opening and closing of the elevator is controlled. In addition, the control signal and the alarm signal can be output simultaneously for multimedia playing in the elevator.
According to one embodiment of the present invention, in step S3, all or a portion of the first image and all or a portion of the second image are uploaded for screening by the user and the screening result is obtained; and the screening result is a false detection or missing detection image in the first image and the second image. In this embodiment, the first image in which the target is detected is an image for performing the elevator control. In the present embodiment, the mobile application and/or the web service can be selectively used to perform the screening operation of the user.
As shown in fig. 1, according to an embodiment of the present invention, the step of updating the target detection model with the screening result in step S4 includes:
s41, obtaining a screening result and an uncertain target image, and counting the data volume of the uncertain target image; in this embodiment, the screening result is an image of false detection or missed detection screened by the user in the foregoing step, and the uncertain target image is an image of an uncertain portion with low or low confidence level in the foregoing step. The continuous collection of image data is performed during the acquisition of these images to achieve data accumulation and to count the amount of data.
S42, if the data volume reaches a first preset value, selectively combining a screening result, an uncertain target image and an initial sample set for offline training of a target detection model to obtain a fine-tuning training set for offline fine-tuning of the target detection model;
in this embodiment, the first preset value may be determined according to a sample data size of an initial sample set used for offline training of the target detection model, for example, set to be 5% of the sample data size of the initial sample set.
In this embodiment, in step S42, the step of selectively combining the screening result, the uncertain target image, and the initial sample set for offline training of the target detection model to obtain a fine-tuning training set for offline fine-tuning of the target detection model includes:
and S421, rechecking the screening result and the data volume of the uncertain target image. In this embodiment, the CVAT method can be used.
S422, extracting a certain amount of sample images in an initial training set of the initial sample set for the offline training target detection model according to a preset proportion to serve as a first training set;
s423, dividing the screening result and the uncertain target image into a second training set and a first testing set according to the original training testing set proportion in the initial sample set;
and S424, combining the first training set and the second training set as a fine tuning training set, and combining an initial test set and a first test set which are used for testing in the initial sample set as a fine tuning test set.
S43, performing off-line fine adjustment training on the target detection model based on the fine adjustment training set to obtain a fine adjustment target detection model;
in this embodiment, in step S43, in the step of performing offline fine-tuning training on the target detection model based on the fine-tuning training set to obtain the fine-tuning target detection model, the fine-tuning training set of the merged data is used to perform fine-tuning on the original model parameters, where a smaller learning rate is required and a smaller number of rounds of training is required, including:
s431, performing offline fine tuning training on the target detection model by adopting an Adam optimizer; wherein the initial learning rate is set to 1e-5 and the minimum learning rate is set to 1 e-6;
and S432, training 5 rounds based on the fine tuning training set, wherein a Cosine learning rate adjusting method is used for adjusting the learning rate in the training process.
And S44, verifying the fine-tuning target detection model, and updating the target detection model running on line based on the parameters of the fine-tuning target detection model if the verification result meets the preset requirement. In this embodiment, the preset requirement is that the average precision of the fine-tuning target detection model on the fine-tuning test set mAP (mean average precision) meets a preset threshold (for example, the error between the average precision and the original target detection model mAP is selected to be within 1%);
and if the average precision of the fine-tuning target detection model on the fine-tuning test set meets a preset threshold, replacing the parameters of the target detection model running on the line with the parameters of the fine-tuning target detection model.
According to an embodiment of the present invention, s44, verifying the fine-tuning target detection model, and if the verification result meets a preset requirement, updating the target detection model running on the line based on the parameters of the fine-tuning target detection model, further includes:
updating an initial sample set, comprising:
randomly extracting sample images with the same number as that of the first training set in the fine tuning training set, and putting the sample images into the initial sample set, so that the number of the original training set is not changed, and difficult samples are added in the original training set;
half of the sample images are randomly drawn in the first test set, replacing an equal number of images in the initial test set.
According to an embodiment of the present invention, in the step of verifying the fine-tuning target detection model in step S44, if the verification result does not satisfy the preset requirement, part (e.g., 50%) of the previously acquired screening results and uncertain target images are deleted, and step S41 is executed again to continue acquiring new screening results and uncertain target images for data accumulation until the data amount reaches the first preset value again, and steps S42 to S44 are executed.
As shown in fig. 1, according to an embodiment of the present invention, an elevator early warning control system based on adaptive learning of the present invention includes: the system comprises a data acquisition module, front-end equipment, an elevator controller, a user feedback module and a cloud server;
the data acquisition module is used for acquiring video image information of the elevator;
the front-end equipment transmits video image information to the cloud server and receives a detection result sent by the cloud server;
the elevator controller is used for receiving the detection result and controlling and giving an alarm to the elevator;
the user feedback module is used for receiving the detection result uploaded by the front-end equipment and screening the detection result;
and the cloud server is used for receiving the screening result uploaded by the user feedback module and carrying out target detection on the video image information to output a detection result.
In this embodiment, the cloud server includes: an incremental image data module and an adaptive learning module. In this embodiment, the incremental image data module is configured to store an initial sample set and a screening result for offline training of the target detection model; the self-adaptive learning module is used for performing offline fine-tuning training on the target detection model.
The foregoing is merely exemplary of particular aspects of the present invention and devices and structures not specifically described herein are understood to be those of ordinary skill in the art and are intended to be implemented in such conventional ways.
The above description is only one embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (11)
1. An elevator early warning control method based on adaptive learning comprises the following steps:
s1, collecting video image information of an elevator, and performing online target detection on the video image information by adopting a target detection model to obtain a detection result; wherein the detection result comprises: a first image in which an object is detected and a second image in which an object is not detected;
s2, controlling and giving an alarm to the elevator based on the detection result;
s3, uploading all or part of the first image and all or part of the second image for screening by a user and acquiring a screening result; wherein the screening result is a false detection or missing detection image in the first image and the second image;
and S4, updating the target detection model by using the screening result.
2. The elevator early warning control method according to claim 1, wherein in step S1, acquiring video image information of an elevator, performing online target detection on the video image information by using a target detection model, and acquiring a detection result, the method comprises:
s11, acquiring the first image with the target and the second image without the target in the video image information based on a first threshold value;
and S12, acquiring an uncertain target image in the first image based on a second threshold value.
3. The elevator early warning control method according to claim 1, wherein the step of updating the target detection model using the screening result in step S4 includes:
s41, acquiring the screening result and the uncertain target image, and counting the data volume of the uncertain target image;
s42, if the data volume reaches a first preset value, selectively combining the screening result, the uncertain target image and an initial sample set for off-line training of the target detection model to obtain a fine-tuning training set for off-line fine tuning of the target detection model;
s43, performing off-line fine tuning training on the target detection model based on the fine tuning training set to obtain a fine tuning target detection model;
and S44, verifying the fine-tuning target detection model, and updating the target detection model running on line based on the parameters of the fine-tuning target detection model if the verification result meets the preset requirement.
4. The elevator early warning control method according to claim 1, wherein in the step of verifying the fine-tuning target detection model in step S44, if the verification result does not meet a preset requirement, part of the previously acquired screening result and the uncertain target image are deleted, and step S41 is executed again to continue acquiring new screening result and uncertain target image for data accumulation until the data amount reaches the first preset value again, and steps S42 to S44 are executed.
5. The elevator early warning control method according to any one of claims 1 to 4, wherein in step S42, the step of selectively combining the screening result, the uncertain target image and an initial sample set for off-line training of the target detection model to obtain a fine-tuning training set for off-line fine-tuning of the target detection model comprises:
s421, rechecking the screening result and the data volume of the uncertain target image;
s422, extracting a certain amount of sample images in an initial training set of the initial sample set for off-line training the target detection model according to a preset proportion to serve as a first training set;
s423, dividing the screening result and the uncertain target image into a second training set and a first testing set according to the original training testing set proportion in the initial sample set;
and S424, combining the first training set and the second training set as the fine tuning training set, and combining the initial test set used for testing in the initial sample set and the first test set as the fine tuning test set.
6. The method as claimed in claim 5, wherein the step of performing offline fine-tuning training on the target detection model based on the fine-tuning training set to obtain the fine-tuning target detection model in step S43 includes:
s431, performing offline fine tuning training on the target detection model by adopting an Adam optimizer; wherein the initial learning rate is set to 1e-5 and the minimum learning rate is set to 1 e-6;
and S432, training 5 rounds based on the fine tuning training set, wherein a Cosine learning rate adjusting method is used for adjusting the learning rate in the training process.
7. The elevator early warning control method according to claim 6, wherein in step S44, the fine-tuning target detection model is verified, and if the verification result meets a preset requirement, the target detection model running on line is updated based on the parameters of the fine-tuning target detection model, wherein the preset requirement is that the average precision of the fine-tuning target detection model on the fine-tuning test set meets a preset threshold;
and if the average precision of the fine-tuning target detection model on the fine-tuning test set meets a preset threshold, replacing the parameters of the target detection model running on the line with the parameters of the fine-tuning target detection model.
8. The elevator early warning control method according to claim 5, wherein S44, the step of verifying the fine-tuning target detection model, and if the verification result meets a preset requirement, updating the target detection model running on line based on the parameters of the fine-tuning target detection model further comprises:
updating the initial sample set, comprising:
randomly extracting sample images with the same quantity as the first training set in the fine tuning training set, and putting the sample images into the initial sample set;
randomly sampling half of the sample images in the first test set to replace an equal number of images in the initial test set.
9. The elevator early warning control method according to claim 1, wherein in step S1, the target detection model is yolov5 model, which outputs corresponding detection result for each input image;
the detection result is expressed as: n x (c, s, x)1,y1,x2,y2)；
Wherein N is the number of detected targets, c is the category of the detected targets, and s is the confidence of the detected targets; (x)1,y1,x2,y2) Coordinate positions of the upper left corner and the lower right corner of the rectangular frame in the image are detected.
10. A system using the adaptive learning-based elevator early warning control method according to any one of claims 1 to 9, comprising: the system comprises a data acquisition module, front-end equipment, an elevator controller, a user feedback module and a cloud server;
the data acquisition module is used for acquiring video image information of the elevator;
the front-end equipment transmits the video image information to the cloud server and receives a detection result sent by the cloud server;
the elevator controller is used for receiving the detection result and controlling and giving an alarm to the elevator;
the user feedback module is used for receiving the detection result uploaded by the front-end equipment and screening the detection result;
and the cloud server is used for receiving the screening result uploaded by the user feedback module, carrying out target detection on the video image information and outputting the detection result.
11. The system of claim 10, wherein the cloud server comprises: an incremental image data module and an adaptive learning module;
the incremental image data module is used for storing an initial sample set and the screening result for off-line training of the target detection model;
the self-adaptive learning module is used for performing offline fine-tuning training on the target detection model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110716814.5A CN113378767A (en) | 2021-06-28 | 2021-06-28 | Elevator early warning control method and system based on adaptive learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110716814.5A CN113378767A (en) | 2021-06-28 | 2021-06-28 | Elevator early warning control method and system based on adaptive learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113378767A true CN113378767A (en) | 2021-09-10 |
Family
ID=77579389
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110716814.5A Pending CN113378767A (en) | 2021-06-28 | 2021-06-28 | Elevator early warning control method and system based on adaptive learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113378767A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110837870A (en) * | 2019-11-12 | 2020-02-25 | 东南大学 | Sonar image target identification method based on active learning |
| CN111524119A (en) * | 2020-04-22 | 2020-08-11 | 征图新视(江苏)科技股份有限公司 | Two-dimensional code defect detection method based on deep learning |
| CN111666993A (en) * | 2020-05-28 | 2020-09-15 | 平安科技（深圳）有限公司 | Medical image sample screening method and device, computer equipment and storage medium |
| CN112079213A (en) * | 2020-08-24 | 2020-12-15 | 浙江新再灵科技股份有限公司 | Elevator entry control method and elevator entry control system |
| CN112149721A (en) * | 2020-09-10 | 2020-12-29 | 南京大学 | Target detection method for reducing labeling requirements based on active learning |
| CN112966772A (en) * | 2021-03-23 | 2021-06-15 | 之江实验室 | Multi-person online image semi-automatic labeling method and system |
-
2021
- 2021-06-28 CN CN202110716814.5A patent/CN113378767A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110837870A (en) * | 2019-11-12 | 2020-02-25 | 东南大学 | Sonar image target identification method based on active learning |
| CN111524119A (en) * | 2020-04-22 | 2020-08-11 | 征图新视(江苏)科技股份有限公司 | Two-dimensional code defect detection method based on deep learning |
| CN111666993A (en) * | 2020-05-28 | 2020-09-15 | 平安科技（深圳）有限公司 | Medical image sample screening method and device, computer equipment and storage medium |
| CN112079213A (en) * | 2020-08-24 | 2020-12-15 | 浙江新再灵科技股份有限公司 | Elevator entry control method and elevator entry control system |
| CN112149721A (en) * | 2020-09-10 | 2020-12-29 | 南京大学 | Target detection method for reducing labeling requirements based on active learning |
| CN112966772A (en) * | 2021-03-23 | 2021-06-15 | 之江实验室 | Multi-person online image semi-automatic labeling method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112883819B (en) | Multi-target tracking method, device, system and computer readable storage medium | |
| CN104063883B (en) | A kind of monitor video abstraction generating method being combined based on object and key frame | |
| CN104538041B (en) | abnormal sound detection method and system | |
| CN106997629A (en) | Access control method, apparatus and system | |
| CN107679471B (en) | Indoor personnel air post detection method based on video monitoring platform | |
| CN105469105A (en) | Cigarette smoke detection method based on video monitoring | |
| CN108229333A (en) | For identifying the method for the event in sport video | |
| CN105791774A (en) | Surveillance video transmission method based on video content analysis | |
| CN102348101A (en) | Examination room intelligence monitoring system and method thereof | |
| CN101727672A (en) | Method for detecting, tracking and identifying object abandoning/stealing event | |
| CN111667501A (en) | Target tracking method and device, computing equipment and storage medium | |
| CN111460940B (en) | Method and system for studying and judging stranger footfall points | |
| CN101299275A (en) | Method and device for detecting target as well as monitoring system | |
| CN105141923B (en) | A kind of video concentration method and device | |
| GB2404511A (en) | Image capture device having a learning function | |
| CN110688884B (en) | Passenger flow statistical method and device | |
| CN111209781B (en) | Method and device for counting indoor people | |
| DE102017130194A1 (en) | Learning method, corresponding system, corresponding device and corresponding computer program product | |
| CN112016731B (en) | Queuing time prediction method and device and electronic equipment | |
| CN107016361A (en) | Recognition methods and device based on video analysis | |
| US8238603B1 (en) | Systems and methods for multi-pass adaptive people counting | |
| CN112287753A (en) | System for improving face recognition precision based on machine learning and algorithm thereof | |
| CN110171754B (en) | Intelligent volume adjusting method and system for elevator advertising screen | |
| CN112087603A (en) | Intelligent examination room supervision method | |
| CN106781167A (en) | The method and apparatus of monitoring object motion state |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210910 |
|
| RJ01 | Rejection of invention patent application after publication |