CN109583355B - People flow counting device and method based on boundary selection - Google Patents
People flow counting device and method based on boundary selection Download PDFInfo
- Publication number
- CN109583355B CN109583355B CN201811406445.4A CN201811406445A CN109583355B CN 109583355 B CN109583355 B CN 109583355B CN 201811406445 A CN201811406445 A CN 201811406445A CN 109583355 B CN109583355 B CN 109583355B
- Authority
- CN
- China
- Prior art keywords
- yolo
- boundary
- neural network
- people
- detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000013528 artificial neural network Methods 0.000 claims abstract description 58
- 238000001514 detection method Methods 0.000 claims abstract description 45
- 230000006835 compression Effects 0.000 claims abstract description 14
- 238000007906 compression Methods 0.000 claims abstract description 14
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 8
- 230000006399 behavior Effects 0.000 claims description 5
- 238000013135 deep learning Methods 0.000 abstract description 3
- 238000012545 processing Methods 0.000 abstract description 2
- 238000013527 convolutional neural network Methods 0.000 description 7
- 238000003062 neural network model Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 230000000052 comparative effect Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/53—Recognition of crowd images, e.g. recognition of crowd congestion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a device and a method for counting people flow based on boundary selection, and belongs to the field of deep learning and image processing. The invention improves the YOLO neural network, namely the division of the YOLO units is increased from 7 x 7 to 9 x 9, the detection number of each unit is increased to 3, the 16 th, 18 th and 24 th convolution layers in the YOLO-PC neural network are respectively replaced by the Fire module in the SqueezeNet, the number of convolution kernels of a compression part in the Fire module is reduced from 128 to 96, the network is retrained, a new S-YOLO-PC neural network can be obtained, the S-YOLO-PC neural network is used for carrying out the people flow statistics of boundary selection, the accuracy of the S-YOLO-PC neural network is improved under the condition that the model is greatly reduced, and the S-YOLO-PC neural network can be used for people flow detection in various occasions.
Description
Technical Field
The invention relates to a device and a method for counting people flow based on boundary selection, belonging to the field of deep learning and image processing.
Background
In a video monitoring picture, pedestrians are very important detection targets, and the tasks of pedestrian detection, crowd density estimation, people flow statistics and the like are all key components of intelligent security and intelligent buildings. The variable and complex backgrounds in the video bring difficulties to the detection of pedestrians and other types of objects in a monitoring picture and the effective separation of pedestrians and backgrounds of pedestrians. The work often has the problems of inaccurate detection, inaccurate counting, long detection and counting result delay, large occupied storage space of the depth model, unfavorable transmission and downloading and the like in a specific scene.
In recent years, the deep learning technology raises the wave of artificial intelligence, and in the fields of image classification and target detection, the deep convolutional neural network makes a series of major breakthroughs, so that the accuracy is greatly improved. However, the related tasks of robot control and automatic driving technology are not only to seek accuracy, but also to speed, and must depend on a low-delay system. In recent years, R-CNN, Fast R-CNN and Faster R-CNN represent the most advanced level of target detection, and YOLO (You Only Look one) is the method with the best real-time performance.
The YOLO is an advanced real-time target detection algorithm, the accuracy rate is high, but many problems are still encountered in the actual application environment, for example, the YOLO occupies a large storage space, is not beneficial to transmission, downloading and detection in the actual application, and severely limits the actual application of the YOLO, and the YOLO 2 updated after 2017 and the YOLO 3 later use the idea of faster-CNN for reference, and the detection mechanism of the anchor is introduced, so that the new YOLO neural network cannot use a boundary selection method to count pedestrians.
Therefore, a method for people stream statistics with small storage space and high real-time performance is needed.
The SqueezeNet is a network structure of a small model that is compressed in a lossy manner using an existing Convolutional Neural Network (CNN) model. And training the network model by using a small amount of parameters to realize the compression of the model. The structure of the Fire module is divided into a compression part and an expansion part, and the compression part and the expansion part are connected to form the tissue convolution filter in the Fire module. The usual SqueezeNet starts with a separate convolutional layer (conv1), then 8 Fire modules, and finally a final translation layer (conv 10).
Disclosure of Invention
In order to solve the problems, the invention provides a people flow statistics method based on boundary selection, and the invention improves on the basis of a Yolov1 neural network, so that the storage space of the people flow statistics method is obviously reduced, and people flow statistics with higher real-time performance is realized without influencing the accuracy of the people flow statistics method.
The technical scheme of the invention is as follows: a method of people flow statistics based on boundary selection, the method comprising the steps of:
step 1: setting the height and the angle of the camera so that the collected pictures can cover the area of the people flow to be measured, and collecting the people flow pictures through the camera;
step 2: setting the detection confidence coefficient of the S-YOLO-PC neural network through equipment containing a GPU;
and step 3: reading an image collected by a camera through an S-YOLO-PC neural network;
and 4, step 4: setting the boundary of an S-YOLO-PC neural network through a device containing a GPU and detecting pedestrian behaviors;
and 5: counting people flow information by an S-YOLO-PC neural network;
step 6: outputting result information of people flow statistics to a computer screen or a screen of a camera in real time through equipment containing a GPU;
the S-YOLO-PC neural network is an improved YOLO neural network: increasing the division of the YOLO units from 7 x 7 to 9 x 9, increasing the detection number of each unit to 3 to obtain a YOLO-PC neural network, respectively replacing 16 th, 18 th and 24 th convolution layers in the YOLO-PC neural network by a Fire module in the SqueezeNet, reducing the number of convolution kernels of a compression part in the Fire module from 128 to 96, and retraining the network to obtain the S-YOLO-PC neural network.
In one embodiment of the present invention, the confidence of detection in step 2 is 0.2-0.4.
In one embodiment of the invention, the training network only trains the class of targets "people".
In one embodiment of the present invention, the boundary is one or more grid cells selected from 243 YOLO cells as a region boundary, and different boundaries are selected according to the actual situation of people, when people turn left from a certain place, the boundary of the left region of the video is selected, and the values of the boundaries are values in 81-89, 108-; when people turn right from a certain place, the value of the boundary is the value in 99-107, 126-; when people go straight, the boundary values are the values in 90-98, 117, 125 or 144, 152.
In an embodiment of the present invention, the statistical people flow information specifically includes: the pedestrian flow statistic value is S/n, wherein S represents the number of detection frames detected at the time t, and n is the number of times that the pedestrian is repeatedly detected in the set confidence coefficient and the selected boundary area.
In one embodiment of the invention, when the detection confidence is 0.2, the number of times that the pedestrian is repeatedly detected is 18, and the people flow statistic is S/18; when the detection confidence coefficient is 0.3, the number of times that the pedestrian is repeatedly detected is 16, and the pedestrian flow statistic value is S/16; when the detection confidence is 0.4, the number of times the pedestrian is repeatedly detected is 13, and the pedestrian flow statistic is S/13.
In an embodiment of the present invention, the signal output in step 6 is to directly display the number of real-time people streams on a camera or a device including a GPU for acquiring images, or to store, continuously accumulate and update real-time information, and then display the real-time information through other interfaces.
The invention also provides a device for people stream statistics based on boundary selection,
the device comprises an image acquisition device, a calculation module and an output module, wherein the calculation module comprises a calculation network and hardware equipment, the image acquisition module is used for acquiring data, the calculation network runs on the hardware equipment to read acquired images and identify and count pedestrian behaviors, and then the pedestrian flow statistical information is output through the hardware equipment or the image acquisition device;
wherein the hardware device is a device including a GPU, the computing network is an S-YOLO-PC neural network, and the S-YOLO-PC neural network is an improved YOLO neural network: increasing the division of the YOLO units from 7 x 7 to 9 x 9, increasing the detection number of each unit to 3 to obtain a YOLO-PC neural network, respectively replacing 16 th, 18 th and 24 th convolution layers in the YOLO-PC neural network by a Fire module in the SqueezeNet, reducing the number of convolution kernels of a compression part in the Fire module from 128 to 96, and retraining the network to obtain the S-YOLO-PC neural network.
The invention has the beneficial technical effects that:
(1) the S-YOLO-PC neural network increases the division of the original YOLO unit from 7 multiplied by 7 to 9 multiplied by 9, increases the detection number of each unit to 3, can obviously improve the average precision of pedestrian detection, and improves the average precision by 14.5 percent compared with the original YOLO neural network;
(2) according to the invention, the Fire module in the SqueezeNet is introduced to replace a specific 3 x 3 convolution layer in the original YOLO neural network, and the number of convolution kernels of a compression part in the Fire module is reduced from 128 to 96, so that the storage size of the neural network can be obviously compressed and reduced by 36.5% compared with the size of the original YOLO neural network under the condition of ensuring the accuracy, the requirement of a network mechanism on the GPU performance is greatly reduced, and the network mechanism has a wider application scene.
(3) The present invention uses a method of defining a boundary so that the S-YOLO-PC can ignore some meaningless disturbances, such as people in a billboard and irrelevant backs, and thus the S-YOLO-PC can more accurately detect the flow of pedestrians.
Drawings
FIG. 1: a flow diagram of a method of people flow statistics based on boundary selection.
FIG. 2: graph of experimental results when the left border was determined.
Detailed Description
AP: namely, Average Precision, the calculation formula is as follows:
where P (i) refers to the precision at which the threshold i is given, and Δ r (i) refers to the recall change value between k and k-1.
The training and testing data set is derived from PASCAL VOC (The pattern analysis, statistical and computational Visual Object Classes Project) and is divided into two parts of VOC2007 and VOC2012, The invention only trains The human class, 5011 picture of VOC2007 data set and 11540 pictures of VOC2012 data set are used as training data, The training data is 16551, The testing set is VOC2007 testing set, and The testing set is 4952 testing pictures.
Example 1
The technical scheme of the invention is as follows: a method of people flow statistics based on boundary selection, the method comprising the steps of:
step 1: setting the height and the angle of the camera so that the collected pictures can cover the area of the people flow to be measured, and collecting the people flow pictures through the camera;
step 2: setting the detection confidence coefficient of the S-YOLO-PC neural network through a computer;
and step 3: increasing the division of a YOLO unit from 7 × 7 to 9 × 9, increasing the detection number of each unit to 3 to obtain a YOLO-PC neural network, respectively replacing 16 th, 18 th and 24 th convolution layers in the YOLO-PC neural network by a Fire module in the SqueezeNet, reducing the number of convolution kernels of a compression part in the Fire module from 128 to 96, re-training the network aiming at 2007+2012 data sets, training only a target of 'human' to obtain an S-YOLO-PC neural network, and reading an image collected by a camera;
and 4, step 4: the boundary of the S-YOLO-PC neural network is set by a computer: when a person turns left from a certain place, selecting the boundary of the left area of the video, the value 113 of the boundary; when a person turns right from a certain place, the value of the boundary is 129; when a person walks straight, the boundary value is 121, and pedestrian behavior is detected;
and 5: the S-YOLO-PC neural network counts people flow information, when the confidence coefficient is 0.2, the number of times that the pedestrian is repeatedly detected is 18, and the corresponding people flow statistic value is S/18, wherein the S value represents the number of detection frames detected at the time t;
step 6: and outputting and displaying the result information of people flow statistics, including the current number of people and the confidence value, on a computer in real time.
And carrying out the same people flow detection and statistics by using the YOLO-PC neural network and the original YOLO neural network according to the same method, and comparing three different neural networks and data for people flow detection and statistics.
(1) Storage size of neural network
Comparing the memories of the three neural network models, the results are shown in table 1, and it can be seen that the storage size of the S-YOLO-PC neural network is reduced by 36.5% compared with the original YOLO and is reduced by 9.1% compared with the YOLO-PC.
TABLE 1 storage size of different neural network models
The S-YOLO-PC neural network replaces the convolution layer in the YOLO-PC with the Fire module, and for the input channel number of 512 or 1024, the result is shown in table 2, after the replacement, the obtained parameters are greatly reduced compared with the original convolution layer, for example, when the input channel number is 512, the number of the parameters is reduced by about 84.7% compared with the number of the original convolution layer, and further the number of the convolution kernels in the Fire module is reduced from 128 to 96, the parameters are continuously reduced by 25%, and the parameters are further reduced, which shows that the corresponding parameters are greatly reduced by replacing the convolution layer in the YOLO with the Fire module. However, the accuracy is greatly reduced due to the large reduction of the parameters, so that a specific convolutional layer needs to be searched for the replacement of the Fire module, and the accuracy can be kept unchanged.
TABLE 2 differences of s1Different ciResulting change in parameters
Wherein s is1Number of convolution kernels for Fire module, ciIs the input channel number.
(2) Average precision AP
Aiming at three different neural networks YOLO, YOLO-PC and S-YOLO-PC, people flow detection is carried out according to the same method, the average precision is shown in Table 3, the precision of S-YOLO-PC is obviously improved by about 14.5% compared with the original YOLO, and the precision of the YOLO-PC and the S-YOLO-PC is almost unchanged, so that the invention obviously reduces the model storage size (by about 9%) by carrying out Fire model replacement on 16 th, 18 th and 24 th convolution layers in the YOLO-PC and adjusting the number of convolution kernels to be reduced from 128 to 96, but can ensure that the precision is almost consistent.
TABLE 3 average accuracy of pedestrian detection by different methods
(3) Confidence value
Three groups of experiments of 4-minute real-time videos are set by using different confidence levels (namely 0.2, 0.3 and 0.4) and are tested, and the results are shown in table 4, S-YOLO-PC obtains more detection frames than YOLO detection, and the average confidence value is more than 50%, which is obviously higher than that of the original YOLO model, but the average confidence value is reduced compared with YOLO-PC, which is caused by the compression of data by a Fire module, but the influence on the accuracy rate is small on the whole.
TABLE 4 comparison of number of detection boxes and average confidence level under different thresholds for different methods
Comparative example 1
Replacing the 15 th, 17 th and 24 th layers with 1024 input in the YOLO neural network by the Fire module in the SqueezeNet respectively, the storage size of the neural network model becomes large and is 738.1M, and the accuracy rate becomes small and is 71%.
Compared with the conventional method that the Fire module replaces 16 th and 18 th layers, the method replaces the layers with 512 input convolution kernels, has smaller model and higher accuracy compared with the method that all the layers with 1024 input kernels are replaced, and can be seen that the replacement position of the Fire module is required to be specific to achieve the effect of the method.
Comparative example 2
The 16 th layer, the 18 th layer and the 19 th layer in the YOLO neural network are respectively replaced by the Fire module in the SqueezeNet, the 19 th layer is also a convolution layer with 1024 input convolution kernels, but the storage size of the neural network model obtained after replacement is unchanged, but the accuracy is reduced.
Comparative example 3
When the number of convolution kernels of the compression part in the Fire module is kept to be 128, as can be seen from the above table 2, when the number of convolution kernels of the compression part in the Fire module is 128, the number of parameters is obviously increased compared with that when the number of convolution kernels of the compression part is 96, the storage size of the neural network model is obviously increased, and the accuracy rate is not obviously improved.
Although the present invention has been described with reference to the preferred embodiments, it should be understood that various changes and modifications can be made therein by those skilled in the art without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (9)
1. A method for people flow statistics based on boundary selection, the method comprising the steps of:
step 1: setting the height and the angle of the camera so that the collected pictures can cover the area of the people flow to be measured, and collecting the people flow pictures through the camera;
step 2: setting the detection confidence coefficient of the S-YOLO-PC neural network through equipment containing a GPU;
and step 3: reading an image collected by a camera through an S-YOLO-PC neural network;
and 4, step 4: setting the boundary of an S-YOLO-PC neural network through a device containing a GPU and detecting pedestrian behaviors;
and 5: counting people flow information by an S-YOLO-PC neural network;
step 6: outputting result information of people flow statistics to a computer screen or a screen of a camera in real time through equipment containing a GPU;
the S-YOLO-PC neural network is an improved YOLO neural network: increasing the division of the YOLO units from 7 × 7 to 9 × 9, increasing the detection number of each unit to 3, obtaining a YOLO-PC neural network, and obtaining 243 detection areas; respectively replacing 16 th, 18 th and 24 th 3 x 3 convolution layers in the YOLO-PC neural network by a Fire module in the SqueezeNet, reducing the number of convolution kernels of a compression part in the Fire module from 128 to 96, and retraining the network to obtain the S-YOLO-PC neural network;
the boundary is that one or more regions are selected from 243 detection regions as region boundaries, different boundaries are selected according to the actual situation of people, when people turn left from a certain place, the boundary of the left region of the video is selected, and the value of the boundary is 81-89, 108-116 or 135-143; when people turn right from a certain place, the value of the boundary is the value in 99-107, 126-; when people go straight, the boundary values are the values in 90-98, 117, 125 or 144, 152.
2. The method of claim 1, wherein the confidence level of detection in step 2 is 0.2-0.4.
3. The method of claim 1, wherein the training network only trains "people" as a class of targets.
4. A method for people flow statistics based on boundary selection according to claim 3, characterized in that the boundary is one or more selected from 243 detected regions as the region boundary, and different boundaries are selected according to the actual situation of people, when people turn left from a certain place, the boundary of the left region of the video is selected, and the value of the boundary is 113; when a person turns right in a certain place, the value of the boundary is 129; when a person goes somewhere, the value of the boundary is 121.
5. The method for people flow statistics based on boundary selection according to any one of claims 1-4, wherein the statistical people flow information is specifically: the pedestrian flow statistic value is S/n, wherein S represents the number of detection frames detected at the time t, and n is the number of times that the pedestrian is repeatedly detected in the set confidence coefficient and the selected boundary area.
6. The method of claim 5, wherein when the detection confidence is 0.2, the number of times the pedestrian is repeatedly detected is 18, and the people flow statistic value is S/18; when the detection confidence coefficient is 0.3, the number of times that the pedestrian is repeatedly detected is 16, and the pedestrian flow statistic value is S/16; when the detection confidence is 0.4, the number of times the pedestrian is repeatedly detected is 13, and the pedestrian flow statistic is S/13.
7. A device for counting people flow based on boundary selection is characterized by comprising an image acquisition device, a calculation module and an output module, wherein the calculation module comprises a calculation network and hardware equipment, the image acquisition module is used for acquiring data, the calculation network runs on the hardware equipment to read acquired images and identify and count pedestrian behaviors, and then people flow statistical information is output through the hardware equipment or the image acquisition device; wherein,
the hardware device is a device containing a GPU, the computing network is an S-YOLO-PC neural network, and the S-YOLO-PC neural network is an improved YOLO neural network: increasing the division of the YOLO units from 7 × 7 to 9 × 9, increasing the detection number of each unit to 3, obtaining a YOLO-PC neural network, and obtaining 243 detection areas; respectively replacing 16 th, 18 th and 24 th 3 x 3 convolution layers in the YOLO-PC neural network by a Fire module in the SqueezeNet, reducing the number of convolution kernels of a compression part in the Fire module from 128 to 96, and retraining the network to obtain the S-YOLO-PC neural network;
the boundary is that one or more regions are selected from 243 detection regions as region boundaries, different boundaries are selected according to the actual situation of people, when people turn left from a certain place, the boundary of the left region of the video is selected, and the value of the boundary is 81-89, 108-116 or 135-143; when people turn right from a certain place, the value of the boundary is the value in 99-107, 126-; when people go straight, the boundary values are the values in 90-98, 117, 125 or 144, 152.
8. The people flow statistics device based on boundary selection according to claim 7 is applied to people flow detection or security and building fields.
9. The use of the method of people stream statistics based on boundary selection according to any one of claims 1-6 in the field of security.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811406445.4A CN109583355B (en) | 2018-11-23 | 2018-11-23 | People flow counting device and method based on boundary selection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811406445.4A CN109583355B (en) | 2018-11-23 | 2018-11-23 | People flow counting device and method based on boundary selection |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109583355A CN109583355A (en) | 2019-04-05 |
| CN109583355B true CN109583355B (en) | 2020-10-09 |
Family
ID=65924204
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811406445.4A Active CN109583355B (en) | 2018-11-23 | 2018-11-23 | People flow counting device and method based on boundary selection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109583355B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020215241A1 (en) | 2019-04-24 | 2020-10-29 | 江南大学 | Real-time object detection method deployed in platform having limited computing resource |
| CN110110627B (en) * | 2019-04-24 | 2021-03-30 | 江南大学 | Real-time target detection method for computing resource limitation platform deployment |
| CN110222673B (en) * | 2019-06-21 | 2021-04-06 | 杭州宇泛智能科技有限公司 | Passenger flow statistical method based on head detection |
| CN110974179A (en) * | 2019-12-20 | 2020-04-10 | 山东大学齐鲁医院 | Auxiliary diagnosis system for stomach precancer under electronic staining endoscope based on deep learning |
| CN112068792A (en) * | 2020-09-09 | 2020-12-11 | 安徽宇烁光电科技有限公司 | External detection system of display screen |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101222482B1 (en) * | 2011-06-23 | 2013-01-15 | 브이씨에이 테크놀러지 엘티디 | People counter having set up interface and set up method thereof |
| CN106886795B (en) * | 2017-02-17 | 2021-01-15 | 北京一维弦科技有限责任公司 | Object identification method based on salient object in image |
| CN108805070A (en) * | 2018-06-05 | 2018-11-13 | 合肥湛达智能科技有限公司 | A kind of deep learning pedestrian detection method based on built-in terminal |
-
2018
- 2018-11-23 CN CN201811406445.4A patent/CN109583355B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109583355A (en) | 2019-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109583355B (en) | People flow counting device and method based on boundary selection | |
| US20230289979A1 (en) | A method for video moving object detection based on relative statistical characteristics of image pixels | |
| Tang et al. | Fast background subtraction and shadow elimination using improved gaussian mixture model | |
| JP2009533778A (en) | Video segmentation using statistical pixel modeling | |
| CN113408671B (en) | Object identification method and device, chip and electronic equipment | |
| CN111062974A (en) | Method and system for extracting foreground target by removing ghost | |
| Porikli et al. | Object tracking in low-frame-rate video | |
| CN110084837B (en) | Target detection and tracking method based on unmanned aerial vehicle video | |
| Roy et al. | Real-time adaptive histogram min-max bucket (HMMB) model for background subtraction | |
| CN106384359B (en) | Motion target tracking method and TV | |
| CN102340620B (en) | Mahalanobis-distance-based video image background detection method | |
| CN113763427A (en) | Multi-target tracking method based on coarse-fine shielding processing | |
| Liu et al. | Scene background estimation based on temporal median filter with Gaussian filtering | |
| Zhang et al. | Multi-domain collaborative feature representation for robust visual object tracking | |
| CN107368802A (en) | Motion target tracking method based on KCF and human brain memory mechanism | |
| Tao et al. | Real-time detection and tracking of moving object | |
| Dou et al. | Modeling the background and detecting moving objects based on Sift flow | |
| RU173468U1 (en) | DEVICE FOR CASCADE PROCESSING OF THE IMAGE FLOW USING SURFACE NEURAL NETWORKS | |
| Chen | Moving object detection based on background extraction | |
| CN116152699B (en) | Real-time moving target detection method for hydropower plant video monitoring system | |
| CN102509076A (en) | Principal-component-analysis-based video image background detection method | |
| Li et al. | Image object detection algorithm based on improved Gaussian mixture model | |
| CN104182990B (en) | A kind of Realtime sequence images motion target area acquisition methods | |
| Takahara et al. | Making background subtraction robust to various illumination changes | |
| Liu et al. | Self-adaptive visual background extraction with ghost regions elimination |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |