CN108549882A - A kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning - Google Patents
A kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning Download PDFInfo
- Publication number
- CN108549882A CN108549882A CN201810457285.XA CN201810457285A CN108549882A CN 108549882 A CN108549882 A CN 108549882A CN 201810457285 A CN201810457285 A CN 201810457285A CN 108549882 A CN108549882 A CN 108549882A
- Authority
- CN
- China
- Prior art keywords
- gpu
- real
- pedestrian detection
- deep learning
- cam
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The present invention proposes a kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning, includes the following steps：The video acquired by multiple cameras, camera are overlapped emphasis monitoring area and monitor；Selection needs the camera shooting head region detected and obtains the camera video stream under the region；Computer Vision is carried out in GPU processor clusters：Image is cut first, then the image after cutting is detected with trained pedestrian detection model；It monitors the GPU service conditions in GPU processors in real time and is scheduled according to scheduling strategy；Pedestrian area coordinate is calculated, current time stamp is obtained and is sent to subscription client.Video processing and deep learning are combined by a kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning, and use the method for camera overlapping monitoring to improve the accuracy rate of railway mouth pedestrian detection.
Description
Technical field
The present invention relates to video processing, artificial intelligence, deep learnings, and in particular to arrives a kind of based on the real-time of deep learning
Railway mouth multi-cam pedestrian detection method.
Background technology
Pedestrian detection is a very popular research topic in target detection research field in complex scene, same with this
When, with the rise of deep learning, the target detection in complex scene has obtained development largely, accuracy of detection and effect
Rate is also higher and higher, however pedestrian detection field still faces many challenges, under the challenge for especially facing railway choma border complexity.
Deep learning was developing very rapidly in recent years, and application field is also more and more extensive, and convolutional neural networks are as deep learning
A kind of method has very significant effect in fields such as target detection, Text regions.By learning to extract in image automatically
Information characteristics greatly reduce manual intervention, and can extract the effective information feature of high quality, to be raising object
Detection is classified, the accuracy rate of target identification lays a solid foundation.
It is more and more to the research of pedestrian detection in complex scene in recent years, it can substantially be divided into two classes：
Method based on background modeling：It is partitioned into foreground, extracts moving target therein, then further extracts feature,
Discriminant classification；Exist rain, snow, blow, leaf shaking, the occasions such as lights keep flickering, the robustness of this method is not high,
Anti-interference ability is poor.And the model of background modeling method is excessively complicated, it is more sensitive to parameter.
Method based on statistical learning：Pedestrian detection grader is built according to a large amount of training samples.The feature of extraction is general
It includes neural network, SVM, adaboost to have the information such as gray scale, edge, texture, shape, the histogram of gradients of target, grader
Deng.There are following difficult points for this method：The posture of pedestrian, dress ornament are different；Distribution of the feature of extraction in feature space is not
It is enough compact；The performance of grader is affected by training sample；Negative sample when off-line training, which can not be covered, all really answers
The case where with scene；Although the pedestrian detection method based on statistical learning is haveed the shortcomings that many, still there are many people will
Focus on this.It is typical represent be French researcher Dalal 2005 pedestrian's inspections of HOG+SVM for delivering of CVPR
Method of determining and calculating.
In recent years, iron crossing traffic and trade rapidly develop, however paste the monitoring at crossing and management level do not catch up with but through
The fast development of Ji can not effectively manage railway mouth, be susceptible to phenomena such as pedestrian illegally enters, and cause traffic thing
Therefore it can cause casualties, property loss, influence the operational paradigm of railway mouth and cause serious economic loss.In order to ensure
Railway security, the technologies such as pedestrian detection are more and more important.Simultaneously early warning in time can largely evade above-mentioned wind to pedestrian detection
Danger helps the existing effective management of railway cause for gossip.Therefore, it is necessary to a kind of effective pedestrian detecting system, monitoring railway mouth passes in and out
Pedestrian improves railway mouth management level, avoids the occurrence of contingency, can be detected in time when generation pedestrian illegally enters phenomenon
And alarm.
Invention content
In order to solve the problems, such as in complex scene pedestrian detection accuracy rate and less efficient, the object of the present invention is to provide one
Real-time railway mouth multi-cam pedestrian detection method of the kind based on deep learning.This method greatly increases pedestrian detection
Accuracy and efficiency.
A kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning of invention, including：
Step 1：The video flowing of monitoring area is acquired by railway mouth camera；
Step 2：The decoded video streams in GPU processor clusters, and carry out image detection processing；
Step 3：It monitors the GPU service conditions in GPU processors in real time and is scheduled according to scheduling strategy；
Step 4：Pedestrian area coordinate is calculated, current time stamp is obtained and is sent to subscription client；
Further, in the step 1, what is be previously mentioned acquires the video flowing of monitoring area by railway mouth camera,
It specifically includes：Multi-cam overlapping monitoring key area is taken, to improve the accuracy rate of pedestrian detection.
Further, in the step 2, the image detection processing being previously mentioned specifically includes：
First by decoding video stream, two points of overlapped partitionings are carried out to decoded image, are then examined with trained pedestrian
Model is surveyed to be detected the image after segmentation.The training process of wherein pedestrian detection model includes：
By excavating history railway mouth video database, extracting effective video data information and carrying out image preprocessing structure
SSD pedestrian's database；
Pedestrian's database is trained using the parameter of SSD network models and acquiescence, according to intermediate result, to initial
Value, training rate and iterations are constantly adjusted, and obtain preferably SSD ship classifications network model.
The beneficial effects of the invention are as follows：
(1) depth learning technology and video processing technique are combined, with the method for deep learning to answering in railway mouth
Miscellaneous scene carries out pedestrian detection, improves the speed and precision of pedestrian detection；
(2) it uses multi-cam to carry out repeating detection to overlapping region, improves the accuracy rate of pedestrian detection.
Description of the drawings
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
Obtain other attached drawings according to these attached drawings.
Fig. 1 is the overview flow chart of the real-time railway mouth multi-cam pedestrian detection based on deep learning in the present invention；
Fig. 2 is the GPU resource scheduling strategy in GPU processor clusters in the present invention；
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation describes, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
As shown in Figure 1, the present invention is based on the real-time railway mouth multi-cam pedestrian detection methods of deep learning, including with
Under several basic steps：Obtain the video flowing under monitoring area；Computer Vision is carried out in GPU processor clusters；Using GPU
Scheduling strategy carries out GPU scheduling；Final result is sent to subscription client.
The real-time railway mouth multi-cam pedestrian detection method based on deep learning is described in detail below：
As shown in Figure 1, installing high-definition camera or video acquisition device in railway mouth, selection needs the region monitored,
Obtain video flowing all under the region；By decoding video stream, two points of overlapped partitionings are carried out to decoded image, then with instruction
The pedestrian detection model perfected is detected the image after segmentation；The GPU monitored in real time in GPU processor clusters uses feelings
Condition takes scheduling strategy appropriate to carry out Real-Time Scheduling to GPU；Pedestrian area coordinate is calculated, current time stamp is obtained and is sent
To subscription client.
GPU resource dispatch layer is according to scheduling strategy as shown in Fig. 2, current GPU resource service condition is monitored in real time, in GPU
Before processor cluster distribution task, whether excessive, if consumption is excessive, check that GPU is used if first checking for current GPU consumption
Situation list and GPU computing capability lists reselect GPU and receive task.
The real-time railway mouth multi-cam pedestrian detection method based on deep learning of the present invention, by depth learning technology
It is combined with video processing technique, pedestrian detection is carried out to railway mouth with the method for deep learning, improves the speed of pedestrian detection
Degree and precision；Overlapping monitoring is carried out to emphasis monitoring section domain using multi-cam, repeats to detect, improves railway mouth pedestrian detection
Accuracy rate.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
With within principle, any modification, equivalent replacement, improvement and so on should all be included in the protection scope of the present invention god.
Claims (5)
1. a kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning, including：At acquisition video, image
Reason, returns the result GPU resource scheduling, it is characterised in that：
Obtain the video flowing under monitoring area；
Computer Vision is carried out in GPU processor clusters；
GPU scheduling is carried out using GPU scheduling strategies；
Final result is sent to subscription client.
2. a kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning as described in claim 1,
Be characterized in that, it is described obtain monitoring area under video flowing the step of, including：
High-definition camera or video acquisition device are installed in railway mouth, selection needs the region monitored, obtains institute under the region
Some video flowings.
3. a kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning as described in claim 1,
The step of being characterized in that, Computer Vision carried out in the GPU processors cluster, including：
First by decoding video stream, two points of overlapped partitionings are carried out to decoded image, then with trained pedestrian detection mould
Type is detected the image after segmentation.
4. a kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning as described in claim 1,
It is characterized in that, described the step of GPU scheduling is carried out using GPU scheduling strategies, including：
It monitors the GPU service conditions in GPU processor clusters in real time, scheduling strategy appropriate is taken to carry out Real-Time Scheduling to GPU.
5. a kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning as described in claim 1,
It is characterized in that, described the step of final result is sent to subscription client：
Pedestrian area coordinate is calculated, current time stamp is obtained and is sent to subscription client.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810457285.XA CN108549882A (en) | 2018-05-14 | 2018-05-14 | A kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810457285.XA CN108549882A (en) | 2018-05-14 | 2018-05-14 | A kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108549882A true CN108549882A (en) | 2018-09-18 |
Family
ID=63494759
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810457285.XA Pending CN108549882A (en) | 2018-05-14 | 2018-05-14 | A kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108549882A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111915779A (en) * | 2020-07-31 | 2020-11-10 | 浙江大华技术股份有限公司 | Gate control method, device, equipment and medium |
| CN112200007A (en) * | 2020-09-15 | 2021-01-08 | 青岛邃智信息科技有限公司 | License plate detection and identification method under community monitoring scene |
-
2018
- 2018-05-14 CN CN201810457285.XA patent/CN108549882A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111915779A (en) * | 2020-07-31 | 2020-11-10 | 浙江大华技术股份有限公司 | Gate control method, device, equipment and medium |
| CN111915779B (en) * | 2020-07-31 | 2022-04-15 | 浙江大华技术股份有限公司 | Gate control method, device, equipment and medium |
| CN112200007A (en) * | 2020-09-15 | 2021-01-08 | 青岛邃智信息科技有限公司 | License plate detection and identification method under community monitoring scene |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jiang et al. | Insulator fault detection in aerial images based on ensemble learning with multi-level perception | |
| Silva et al. | An extended center-symmetric local binary pattern for background modeling and subtraction in videos | |
| CN108052859B (en) | Abnormal behavior detection method, system and device based on clustering optical flow characteristics | |
| CN103839065B (en) | Extraction method for dynamic crowd gathering characteristics | |
| CN106845415B (en) | Pedestrian fine identification method and device based on deep learning | |
| CN108319926A (en) | A kind of the safety cap wearing detecting system and detection method of building-site | |
| CN105426820B (en) | More people's anomaly detection methods based on safety monitoring video data | |
| CN104200466A (en) | Early warning method and camera | |
| CN112287827A (en) | Complex environment pedestrian mask wearing detection method and system based on intelligent lamp pole | |
| CN112232316B (en) | Crowd gathering detection method and device, electronic equipment and storage medium | |
| Liu et al. | An efficient method for high-speed railway dropper fault detection based on depthwise separable convolution | |
| CN114267082B (en) | Bridge side falling behavior identification method based on depth understanding | |
| Yuan et al. | EcRD: Edge-cloud computing framework for smart road damage detection and warning | |
| CN108256447A (en) | A kind of unmanned plane video analysis method based on deep neural network | |
| CN114882440A (en) | Human head detection method and system | |
| CN110909195A (en) | Picture labeling method and device based on block chain, storage medium and server | |
| Yang et al. | Segmentation of track surface defects based on machine vision and neural networks | |
| Hu et al. | Parallel spatial-temporal convolutional neural networks for anomaly detection and location in crowded scenes | |
| CN108549882A (en) | A kind of real-time railway mouth multi-cam pedestrian detection method based on deep learning | |
| CN109002753A (en) | One kind being based on the cascade large scene monitoring image method for detecting human face of convolutional neural networks | |
| Wang et al. | Automatic detection of indoor occupancy based on improved YOLOv5 model | |
| CN112560880A (en) | Object classification method, object classification apparatus, and computer-readable storage medium | |
| Katsamenis et al. | A Few-Shot Attention Recurrent Residual U-Net for Crack Segmentation | |
| CN112258552A (en) | Pedestrian multi-target tracking method under community monitoring scene | |
| CN112270207A (en) | Smoke and fire detection method in community monitoring scene |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180918 |