CN112686180A - Method for calculating number of personnel in closed space - Google Patents
Method for calculating number of personnel in closed space Download PDFInfo
- Publication number
- CN112686180A CN112686180A CN202011645491.7A CN202011645491A CN112686180A CN 112686180 A CN112686180 A CN 112686180A CN 202011645491 A CN202011645491 A CN 202011645491A CN 112686180 A CN112686180 A CN 112686180A
- Authority
- CN
- China
- Prior art keywords
- people
- training
- data set
- personnel
- network
- 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 27
- 238000012549 training Methods 0.000 claims abstract description 47
- 230000006399 behavior Effects 0.000 claims description 26
- 238000013528 artificial neural network Methods 0.000 claims description 16
- 238000012216 screening Methods 0.000 claims description 8
- 238000010586 diagram Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 230000001629 suppression Effects 0.000 claims description 6
- 238000013145 classification model Methods 0.000 claims description 5
- 230000005764 inhibitory process Effects 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 3
- 230000002401 inhibitory effect Effects 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000003247 decreasing effect Effects 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims 1
- 238000005070 sampling Methods 0.000 claims 1
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 238000002372 labelling Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 239000013589 supplement Substances 0.000 description 2
- 238000001514 detection method Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a method for calculating the number of people in a closed space, which relates to the technical field of computers. The invention has the beneficial effects that: according to the invention, through means of video acquisition, model training, behavior recognition and the like, people in the closed space are automatically counted by combining the recognition result.
Description
Technical Field
The invention relates to the technical field of computers, in particular to a method for calculating the number of people in a closed space.
Background
The floating population, especially the floating population in the operation places such as the network contract room and the hotel, has the characteristics of complicated structure, short-term behavior and the like, relatively speaking, the risk of the epidemic situation is higher, and the method brings great challenge to the current epidemic situation management work. One major measure to reduce the risk of an epidemic is to reduce the gathering of people.
Currently, the number of people in a closed space is mainly judged by manual checking or face recognition counting. But the home check needs to consume a large amount of manpower and material resources and has low efficiency. And the face recognition counting is not accurate enough under the condition that the person generally wears the mask.
Disclosure of Invention
Aiming at the technical problem, the invention provides a method for calculating the number of personnel in a closed space.
The method comprises the following steps that firstly, a Yolov5 neural network is used for training photos with people entering a room and leaving the room, and a people behavior classification model capable of classifying people entering the room or leaving the room is obtained;
secondly, arranging video acquisition equipment at an inlet of a closed space to be monitored;
thirdly, carrying out video acquisition on the personnel entering and exiting conditions of the inlet of the closed space, and judging the entering and exiting conditions of the personnel by combining the model in the first step;
and step four, increasing and decreasing the number of the personnel in the closed space through the judgment result of the step three, so as to obtain a total value of the personnel.
Preferably, the method for establishing the personnel behavior classification model comprises the following steps,
s1, preparing a training data set, wherein the data set is a training data set for a person to return to a house or go out when opening a door;
s2, training a model, and performing model training by combining a neural network with the training data set acquired by the S1;
and S3, observing the loss function, and stopping training when the loss function value output by the training data set converges below a threshold value or reaches a set maximum iteration number to obtain a trained recognition network.
Preferably, in S1, a training data set is prepared, specifically,
s1-1, triggering the video acquisition device by a sensor to record a door opening time point when the door is opened through the video acquisition device, and then intercepting videos of a period of time before and after the time point by taking the time point as a center for storage, wherein the videos are used as video data for training of a personnel behavior analysis model, and the video interception time is preferably 15 seconds before and after the door is opened and is 30 seconds in total;
s1-2, using computer equipment to perform frame extraction processing on the video data obtained in the S1-1, and intercepting a picture at intervals of frames and storing the picture in a computer;
s1-3, manually screening out pictures with person entering and exiting behaviors from the pictures obtained in the S1-2;
s1-4, labeling the people in the picture after the S1-3 screening by using a LabelImg tool, and labeling the people with the house-entering behavior and the people with the out-going behavior in the picture differently; selecting and marking the person with the house-entering behavior in the picture as in, and selecting and marking the person with the house-exiting behavior in the picture as out;
s1-5, making the pictures labeled in the S1-4 into data sets according to certain rules and formats, and randomly dividing the data sets into training data sets and verification data sets according to a certain proportion;
specifically, renaming the picture marked by S1-4 according to the format of a Pascal VOC data set, creating a folder with the name of Annotations for storing an xml data set, creating a folder with the name of ImageSets for storing training data and verification data, and creating a folder with the name of JPEGImages for storing a picture set;
and S1-6, normalizing the data in the data set obtained in the S1-5, and converting the normalized data into a txt file required by the neural network, wherein the txt file comprises the type of the picture label, the coordinate of the center point of the label box and the length and the width of the label box.
Specifically, the xml file in the VOC format is converted into the txt file required by the Yolov5 neural network.
Preferably, the S2 model training method specifically includes,
s2-1, setting training parameters of a Yolov5 neural network, wherein the learning rate is 0.001, the iteration number is 40000, the learning rate variation step size is 20000 and 30000, the learning rate variation factor is 0.1 and 0.1, the batch size is 64, the confidence threshold of the target frame is 0.5, the non-maximum value inhibition threshold is 0.3, the depth of the network is 0.33, the width of the network is 0.5, the width of the input data set is 608, the height of the input data set is 608, and the number of channels is 3;
s2-2, sending the prepared data set in S1 to a network for training;
s2-3, formatting the length and width of a picture input into a Yolov5 neural network into 608 pixels × 608 pixels, and obtaining two feature maps with different sizes through operations such as convolution, upsampling, residual error unit, tensor splicing and the like, wherein the two feature maps are 19 × 19 and 38 × 38 respectively;
s2-4, selecting corresponding prediction box sizes from the feature maps of two different sizes obtained in S2-3 according to the range of the sizes of the regions mapped on the input picture by the feature maps, and selecting boundary boxes of 3 sizes from each feature map;
and S2-5, screening the prediction frame obtained in the S2-4 by using an NMS non-maximum value inhibition method, searching a local maximum value, inhibiting non-maximum value elements, eliminating redundant candidate prediction frames, and finding the best prediction frame position.
Preferably, the bounding boxes of 3 sizes in S2-4 are specifically,
s2-4-1, detecting a large target human body in a short distance by using a feature map with the size of 19 x 19, wherein the input images with the sizes of corresponding prediction frames mapped to 608 x 608 are respectively 142 x 110, 192 x 243 and 459 x 401;
s2-4-2, detecting a medium target human body with a medium distance by a feature map with the size of 38 x 38, wherein the input images with the corresponding prediction box sizes mapped to 608 x 608 are respectively 36 x 75, 76 x 55 and 72 x 146;
preferably, the selected network model is Yolov5s, the target of the scheme is a medium target or a large target, so that the network model Yolov5s with the minimum depth and the minimum width of the feature map in the Yolov5 series is selected, and the model identification speed can be increased.
The bottleeckCSP in the Yolov5 network is replaced by the Ghost bottleeck, and linear convolution is performed again on the basis of a small amount of feature graphs obtained by nonlinear convolution, so that more feature graphs are obtained, redundant features are eliminated, a lighter model is obtained, and the training speed is higher.
Preferably, an Squeeze-and-Excitation attention mechanism module is added into the Yolov5 network, the importance degree of each feature channel is automatically acquired in a learning mode, and then useful features are promoted and other features are suppressed according to the importance degree, wherein the other features are features which are not useful for the current task;
preferably, in S2-5, the non-maximum suppression is performed by using DIoU _ NMS. In conventional NMS, the IoU criterion is often used to suppress redundant detection boxes, where overlapping areas are the only factor, often resulting in false suppression for occlusion cases. The DIoU _ NMS considers not only the overlapping regions but also the center point distance between the two prediction boxes. The identification of objects occluding overlapping by the model can be improved.
IoU are defined as follows:
wherein B isgt1,Bgt2Representing two prediction boxes, IoU represents the intersection ratio of the two prediction boxes.
DIoU is defined as follows:
wherein, bgt1,bgt2Represents the center points of the two prediction frames, respectively, p (b)gt1,bgt2) Typically, the euclidean distance between the two center points is calculated. c represents the diagonal Euclidean distance of the minimum closure area which can contain two prediction frames simultaneously.
When NMS non-maximum value suppression is carried out, all prediction boxes are traversed from large to small according to the confidence degrees of the prediction boxes, and the redundant prediction boxes are deleted by simultaneously considering the intersection ratio of the two prediction boxes and the distance between the central points, so that the error suppression of the shielding target is reduced.
When people enter the house, the people enter a shooting area of a video acquisition device at a house entrance, and the house entering or leaving behaviors of the people are identified by combining the people behavior classification model trained by the method. When a person opens and closes the door, the sensor for detecting the state of the door is triggered, the entering and exiting conditions of the person in the time period are recorded, and the person is stored in the computer.
The technical scheme provided by the embodiment of the invention has the following beneficial effects:
1. the number of people in the house can be automatically calculated without manual participation;
2. people counting is carried out without depending on human faces, and when people wear the mask or other reasons, people still can count the number of people in the house through the judgment of the home and the out behaviors of the people when the face information cannot be collected.
Drawings
Fig. 1 is a Yolov5 network modification diagram according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of a person entering a house according to an embodiment of the present invention.
Fig. 3 is a schematic diagram illustrating a counting process of the personnel in the house according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. Of course, the specific embodiments described herein are merely illustrative of the invention and are not intended to be limiting.
It should be noted that the embodiments and features of the embodiments of the present invention may be combined with each other without conflict.
Example 1
Referring to fig. 1 to 3, the invention provides a method for calculating the number of people in a closed space, which comprises configuring a door with a sensor, arranging a video acquisition device at the side of an entrance of the door close to a handle, configuring a light supplement device in an acquisition area of the video acquisition device, and connecting the sensor and the video acquisition device with a computer with a processor and a memory through a network. According to the invention, through means of video acquisition, model training, behavior recognition and the like, people in the closed space are automatically counted by combining the recognition result.
For example, the network reservation room is used, and the camera of the video acquisition device is a wide-angle camera with a light supplement lamp.
The video acquisition device is arranged at the position, close to the handle, outside the door frame, the height of a lens is 1.55 m, and the lens is perpendicular to the wall surface. The bulb in the equipment installation area is not lower than a 10W energy-saving lamp or a 5W LED lamp. The brightness of the light in the equipment snapshot area is about 500-1000 Lux;
the process of model training is as follows:
s1, preparing a training data set, wherein the data set is a training data set for a person to return to a house or go out when opening a door;
s1-1, triggering the video acquisition device by the sensor to record the door opening time point when the door is opened through the video acquisition device, and then intercepting videos of a period of time before and after the time point by taking the time point as the center for storage, wherein the videos are used as video data for training of a personnel behavior analysis model. The video capturing time is preferably 15 before and after the door is opened, and the total time is 30 seconds;
s1-2, using computer equipment to perform frame extraction processing on the video data obtained in the S1-1, and capturing a picture every 10 frames and storing the picture in a computer;
s1-3, manually screening out pictures with person entering and exiting behaviors from the pictures obtained in the S1-2;
s1-4, labeling the people in the picture after the S1-3 screening by using a LabelImg tool, and labeling the people with the house-entering behavior and the people with the out-going behavior in the picture differently; selecting and marking the person with the house-entering behavior in the picture as in, and selecting and marking the person with the house-exiting behavior in the picture as out;
and S1-5, making the pictures labeled in the S1-4 into a data set according to a certain rule and format, and randomly dividing the data set into a training data set and a verification data set according to a certain proportion. Renaming the pictures according to the format of the Pascal VOC data set, creating a folder with the name of being options for storing an xml data set, creating a folder with the name of being ImageSets for storing training data and verification data, and creating a folder of JPEGImages for storing the picture set;
and S1-6, normalizing the data in the data set obtained in the S1-5 and converting the normalized data into a txt file required by the neural network, wherein the content in the file is the type of picture marking, the coordinate of the center point of a marking frame and the length and the width of the marking frame. S2, training a model, and performing model training by combining a neural network with the training data set acquired by the S1;
the model training adopts a Yolov5 neural network algorithm, and because the attention of the user is mainly a large target, a Yolov5s network with the minimum depth, the minimum width of a characteristic diagram and the fastest speed can be selected;
s2-1, setting training parameters of a Yolov5 neural network, preferably setting the learning rate to be 0.001, the iteration number to be 40000, the learning rate variation step size to be 20000 and 30000, the learning rate variation factor to be 0.1 and 0.1, the batch size to be 64, the target frame confidence threshold to be 0.5, the depth of the network to be 0.33, the width of the network to be 0.50, the width of an input data set to be 608, the height to be 608 and the number of channels to be 3;
s2-2, sending the prepared data set in S1 to a network for training;
the method comprises the steps of replacing a bottleckCSP in a Yolov5 network with a Ghost bottleckP, which is abbreviated as Ghost in figure 1, and performing linear convolution once again on the basis of a small amount of feature graphs obtained by nonlinear convolution to obtain more feature graphs, so that redundant features are eliminated, a lighter model is obtained, and the training speed is higher.
An attention mechanism module of Squeeze-and-Excitation, abbreviated as SELayer in fig. 1, is added to a Yolov5 network, the importance degree of each feature channel is automatically obtained in a learning manner, and then useful features are promoted according to the importance degree and features with little use for the current task are suppressed, so that the performance is improved and the learning speed is accelerated.
And removing the feature map of 76 x 76 in the prediction module in the Yolov5 network, and reducing the sensitivity of the model to small targets.
S2-3, formatting the length and width of a picture input into a Yolov5 neural network into 608 pixels × 608 pixels, and obtaining two feature maps with different sizes through convolution, upsampling, residual error unit and tensor splicing, wherein the two feature maps are 19 × 19 and 38 × 38 respectively;
and S2-4, selecting corresponding prediction box sizes from the feature maps of two different sizes obtained in the step S2-3 according to the range of the sizes of the regions mapped on the input picture by the feature maps, and selecting boundary boxes of 3 sizes from each feature map. The feature map with the size of 19 × 19 detects a large target human body in a short distance, and input images with the corresponding prediction frame sizes mapped to 608 × 608 are 142 × 110, 192 × 243 and 459 × 401 respectively; the feature map of size 38 × 38 detects medium target human bodies at medium distances, and the corresponding prediction box sizes are mapped to input images of 608 × 608 as 36 × 75, 76 × 55, and 72 × 146, respectively.
And S2-5, screening the prediction frame obtained in the S2-4 by using an NMS non-maximum value inhibition method, searching a local maximum value, inhibiting non-maximum value elements, eliminating redundant candidate prediction frames, and finding the best prediction frame position. And carrying out non-maximum suppression by using a DIoU _ NMS mode, and being used for improving the identification of the model to the target with the shielding overlapping.
And S2-6, observing a loss function, and stopping training when a loss function value output by the training data set converges below a threshold value or reaches a set maximum iteration number to obtain a trained recognition network.
The behavior recognition process comprises the following steps:
s3, combining the trained recognition network in S2 with a video acquisition device and a sensor to count the personnel who get in and out of the house:
s3-1, detecting that the front of the person appears in the shooting area of the video acquisition device, and then triggering a sensor for detecting the opening and closing state of the door, so that the detected number of the person is added to the gathered number of the persons in the house.
S3-2, triggering a sensor for detecting the opening and closing state of the door, then detecting that the back of the person is in the shooting area of the video acquisition device, and subtracting the detected number of the person from the gathered number of the person in the house.
The present invention is not limited to the above preferred embodiments, and any modifications, equivalent replacements, improvements, etc. within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (8)
1. A method for calculating the number of people in a closed space is characterized by comprising the following steps,
step one, training photos with people entering a room and leaving a room by using a neural network to obtain a people behavior classification model capable of classifying people entering the room or leaving the room;
secondly, arranging video acquisition equipment at an inlet of a closed space to be monitored;
thirdly, carrying out video acquisition on the personnel entering and exiting conditions of the inlet of the closed space, and judging the entering and exiting conditions of the personnel by combining the model in the first step;
and step four, increasing and decreasing the number of the personnel in the closed space through the judgment result of the step three, so as to obtain a total value of the personnel.
2. The closed space personnel number calculation method of claim 1, wherein the personnel behavior classification model is established by a method comprising,
s1, preparing a training data set, wherein the data set is a training data set for a person to return to a house or go out when opening a door;
s2, training a model, and performing model training by combining a neural network with the training data set acquired by the S1;
and S3, observing the loss function, and stopping training when the loss function value output by the training data set converges below a threshold value or reaches a set maximum iteration number to obtain a trained recognition network.
3. The method for calculating the number of people in an enclosed space according to claim 2, wherein a training data set is prepared in the step S1, specifically,
s1-1, triggering the video acquisition device by a sensor to record a door opening time point when a door is opened through the video acquisition device, and then intercepting videos of a period of time before and after the time point by taking the time point as a center for storage, wherein the videos are used as video data for training of a personnel behavior analysis model;
s1-2, performing frame extraction processing on the video data obtained in the S1-1, and intercepting and storing a picture at intervals of frames;
s1-3, screening out pictures of people entering and exiting behaviors from the pictures obtained in the S1-2;
s1-4, marking the people in the picture screened by the S1-3, and marking the people with the house-entering behavior and the people with the house-exiting behavior in the picture differently;
s1-5, making the pictures labeled in the S1-4 into a data set according to a certain rule and format, and randomly dividing the data set into a training data set and a verification data set according to a proportion;
and S1-6, normalizing the data in the data set obtained in the S1-5, and converting the normalized data into a txt file required by the neural network, wherein the txt file comprises the type of the picture label, the coordinate of the center point of the label box and the length and the width of the label box.
4. The method for calculating the number of people in an enclosed space according to claim 3, wherein the S2 training model method is specifically,
s2-1, setting training parameters of a Yolov5 neural network, learning rate, iteration times, learning rate variation step size, learning rate variation factor, batch size, target frame confidence threshold, non-maximum value inhibition threshold, input width, height and channel number of the network;
s2-2, sending the prepared data set in S1 to a network for training;
s2-3, formatting the length and width of the picture input into a neural network into 608 pixels multiplied by 608 pixels, and obtaining two feature maps with different sizes through convolution, up-sampling, residual error unit and tensor splicing;
s2-4, selecting corresponding prediction box sizes from the feature maps of two different sizes obtained in S2-3 according to the range of the sizes of the regions mapped on the input picture by the feature maps, and selecting boundary boxes of 3 sizes from each feature map;
and S2-5, screening the prediction frame obtained in the S2-4 by using an NMS non-maximum value inhibition method, searching a local maximum value, inhibiting non-maximum value elements, eliminating redundant candidate prediction frames, and finding the best prediction frame position.
5. The method for calculating the number of people in the enclosed space according to claim 4, wherein the bounding boxes of two sizes in S2-4 are,
s2-4-1, detecting a large target human body in a short distance by using the characteristic diagram with the size of 19 multiplied by 19, wherein the input images with the sizes of the corresponding prediction frames mapped to 608 multiplied by 608 are respectively 142 multiplied by 110, 192 multiplied by 243, 459 multiplied by 401;
the feature map of size 38 × 38 of S2-4-2 detects medium target human bodies of medium distance, and the corresponding prediction box sizes are mapped to the input images of 608 × 608 as 36 × 75, 76 × 55, 72 × 146, respectively.
6. The method for calculating the number of people in the closed space according to claim 4, wherein the selected network model is Yolov5 s.
Replacing the bottleeckCSP in the Yolov5 network with the Ghost bottleeck, and performing linear convolution again on the basis of a feature diagram obtained by a small amount of nonlinear convolution.
7. The method for calculating the number of people in the closed space according to claim 4, wherein an Squeeze-and-Excitation attention mechanism module is added into a Yolov5 network, the importance degree of each feature channel is automatically acquired in a learning mode, and then useful features are promoted and other features are suppressed according to the importance degree, wherein the other features are features which are not useful for the current task.
8. The method for calculating the number of people in the enclosed space according to claim 4, wherein in the step S2-5, the DIoU _ NMS is used for non-maximum suppression, so as to improve the identification of the model to the target with overlapped occlusion.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011645491.7A CN112686180A (en) | 2020-12-29 | 2020-12-29 | Method for calculating number of personnel in closed space |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011645491.7A CN112686180A (en) | 2020-12-29 | 2020-12-29 | Method for calculating number of personnel in closed space |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112686180A true CN112686180A (en) | 2021-04-20 |
Family
ID=75456956
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011645491.7A Pending CN112686180A (en) | 2020-12-29 | 2020-12-29 | Method for calculating number of personnel in closed space |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112686180A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113160062A (en) * | 2021-05-25 | 2021-07-23 | 烟台艾睿光电科技有限公司 | Infrared image target detection method, device, equipment and storage medium |
CN113469073A (en) * | 2021-07-06 | 2021-10-01 | 西安电子科技大学 | SAR image ship detection method and system based on lightweight deep learning |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040179736A1 (en) * | 2001-05-26 | 2004-09-16 | Yin Jia Hong | Automatic classification and/or counting system |
CN1932843A (en) * | 2006-09-11 | 2007-03-21 | 江苏科技大学 | Distributed demographic system based on insertion system |
CN101540892A (en) * | 2009-04-23 | 2009-09-23 | 上海中安电子信息科技有限公司 | Method for people counting in doorway on DSP video gathering device |
CN110008867A (en) * | 2019-03-25 | 2019-07-12 | 五邑大学 | A kind of method for early warning based on personage's abnormal behaviour, device and storage medium |
CN110837795A (en) * | 2019-11-04 | 2020-02-25 | 防灾科技学院 | Teaching condition intelligent monitoring method, device and equipment based on classroom monitoring video |
CN111382708A (en) * | 2020-03-11 | 2020-07-07 | 广东工业大学 | Method and device for detecting burglary behavior of climbing stairs and turning windows in real time |
CN111476205A (en) * | 2020-02-26 | 2020-07-31 | 安徽建筑大学 | Personnel counting method and device based on L STM model |
CN111986228A (en) * | 2020-09-02 | 2020-11-24 | 华侨大学 | Pedestrian tracking method, device and medium based on LSTM model escalator scene |
CN112001228A (en) * | 2020-07-08 | 2020-11-27 | 上海品览数据科技有限公司 | Video monitoring warehouse in-out counting system and method based on deep learning |
-
2020
- 2020-12-29 CN CN202011645491.7A patent/CN112686180A/en active Pending
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040179736A1 (en) * | 2001-05-26 | 2004-09-16 | Yin Jia Hong | Automatic classification and/or counting system |
CN1932843A (en) * | 2006-09-11 | 2007-03-21 | 江苏科技大学 | Distributed demographic system based on insertion system |
CN101540892A (en) * | 2009-04-23 | 2009-09-23 | 上海中安电子信息科技有限公司 | Method for people counting in doorway on DSP video gathering device |
CN110008867A (en) * | 2019-03-25 | 2019-07-12 | 五邑大学 | A kind of method for early warning based on personage's abnormal behaviour, device and storage medium |
CN110837795A (en) * | 2019-11-04 | 2020-02-25 | 防灾科技学院 | Teaching condition intelligent monitoring method, device and equipment based on classroom monitoring video |
CN111476205A (en) * | 2020-02-26 | 2020-07-31 | 安徽建筑大学 | Personnel counting method and device based on L STM model |
CN111382708A (en) * | 2020-03-11 | 2020-07-07 | 广东工业大学 | Method and device for detecting burglary behavior of climbing stairs and turning windows in real time |
CN112001228A (en) * | 2020-07-08 | 2020-11-27 | 上海品览数据科技有限公司 | Video monitoring warehouse in-out counting system and method based on deep learning |
CN111986228A (en) * | 2020-09-02 | 2020-11-24 | 华侨大学 | Pedestrian tracking method, device and medium based on LSTM model escalator scene |
Non-Patent Citations (2)
Title |
---|
YE LI 等: "A Multi-task Joint Framework for Real-time Person Search", ARXIV:2012.06418V1, 11 December 2020 (2020-12-11), pages 5 - 13 * |
平嘉蓉 等: "基于轻量级神经网络的人群计数模型设计", 无线电工程, no. 06, 19 May 2020 (2020-05-19) * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113160062A (en) * | 2021-05-25 | 2021-07-23 | 烟台艾睿光电科技有限公司 | Infrared image target detection method, device, equipment and storage medium |
CN113469073A (en) * | 2021-07-06 | 2021-10-01 | 西安电子科技大学 | SAR image ship detection method and system based on lightweight deep learning |
CN113469073B (en) * | 2021-07-06 | 2024-02-20 | 西安电子科技大学 | SAR image ship detection method and system based on lightweight deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108052925B (en) | Intelligent management method for community personnel files | |
CN111598040B (en) | Construction worker identity recognition and safety helmet wearing detection method and system | |
CN111563557B (en) | Method for detecting target in power cable tunnel | |
CN109711370A (en) | A kind of data anastomosing algorithm based on WIFI detection and face cluster | |
CN109711309B (en) | Method for automatically identifying whether portrait picture is eye-closed | |
CN110688980B (en) | Human body posture classification method based on computer vision | |
CN107491728A (en) | A kind of human face detection method and device based on edge calculations model | |
CN112287827A (en) | Complex environment pedestrian mask wearing detection method and system based on intelligent lamp pole | |
CN112163572A (en) | Method and device for identifying object | |
CN112686180A (en) | Method for calculating number of personnel in closed space | |
CN110827432B (en) | Class attendance checking method and system based on face recognition | |
CN112287753A (en) | System for improving face recognition precision based on machine learning and algorithm thereof | |
CN109117771B (en) | System and method for detecting violence events in image based on anchor nodes | |
CN113255804A (en) | Garbage traceability method and device based on image change detection | |
CN111754669A (en) | College student management system based on face recognition technology | |
Zambanini et al. | Detecting falls at homes using a network of low-resolution cameras | |
CN102867214B (en) | Counting management method for people within area range | |
CN111080827A (en) | Attendance system and method | |
CN111582195B (en) | Construction method of Chinese lip language monosyllabic recognition classifier | |
CN112183287A (en) | People counting method of mobile robot under complex background | |
CN112541403A (en) | Indoor personnel falling detection method utilizing infrared camera | |
CN117372956A (en) | Method and device for detecting state of substation screen cabinet equipment | |
CN112183460A (en) | Method and device for intelligently identifying environmental sanitation | |
CN115346169B (en) | Method and system for detecting sleep post behaviors | |
CN115909144A (en) | Method and system for detecting abnormity of surveillance video based on counterstudy |
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 |