CN108446662A - A kind of pedestrian detection method based on semantic segmentation information - Google Patents
A kind of pedestrian detection method based on semantic segmentation information Download PDFInfo
- Publication number
- CN108446662A CN108446662A CN201810283404.4A CN201810283404A CN108446662A CN 108446662 A CN108446662 A CN 108446662A CN 201810283404 A CN201810283404 A CN 201810283404A CN 108446662 A CN108446662 A CN 108446662A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- network
- indicate
- training
- semantic segmentation
- 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
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- 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)
- 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)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of pedestrian detection method based on semantic segmentation information, it is related to the pedestrian detection method field based on neural network；It includes 1：Original RGB image input core network, corresponding semantic segmentation image in training set sample are input to branching networks, and the loss function that overall network is arranged completes training；2：The core network that original RGB image input in test set sample is completed to training carries out convolution feature extraction generation multilayer feature figure；3：The Area generation network progress pedestrian candidate frame extraction that multilayer feature figure is inputted to completion training generates pedestrian candidate region；4：It completes to export the detection result image for including pedestrian position encirclement frame after trained classification Recurrent networks are classified and positioned to pedestrian candidate region；It solves the problems, such as that existing pedestrian detection causes accuracy of detection low in low resolution pedestrian with the differentiation of background difficulty, improves the precision of pedestrian detection in the case of low resolution.
Description
Technical field
It is especially a kind of based on semantic segmentation information the present invention relates to the pedestrian detection method field based on neural network
Pedestrian detection method.
Background technology
Pedestrian detection technology is a most basic and common target detection technique in practical applications, it is human body behavior
Analysis, Gait Recognition, intelligent video monitoring and automatic Pilot technology basis.It is emerging with convolutional neural networks in recent years
It rises, huge advance is had been achieved in object detection field, but in pedestrian's detection field, still there is two big challenges：
One, it is detected compared to general target, pedestrian target and background are more difficult to distinguish.Such as in low resolution,
Pedestrian target and traffic lights, column shape target have a closely similar surface characteristics, the pixel distribution of pedestrian and background
Pixel distribution very close to.
Two, it how to be accurately positioned each pedestrian target, in practical applications, usually will appear intensive crowd's scene,
Detector can not be accurately positioned pedestrian target to generate flase drop and missing inspection.In convolutional neural networks, convolutional layer and pond layer
On the one hand high-layer semantic information is generated, the boundary of adjacent target has on the other hand also been obscured, has caused flase drop and missing inspection serious；Therefore
Need a kind of pedestrian detection method that can realize the method accurately detected in low resolution.
Invention content
It is an object of the invention to：The present invention provides a kind of pedestrian detection methods based on semantic segmentation information, solve
Existing pedestrian detection leads to the differentiation of background difficulty that accuracy of detection is low, is based on convolutional Neural net in low resolution pedestrian
Network is divided using semantic information obscures the problem of adjacent target boundary leads to missing inspection and flase drop.
The technical solution adopted by the present invention is as follows：
A kind of pedestrian detection method based on semantic segmentation information, includes the following steps
Step 1：By core network, the corresponding semanteme point in original RGB image input overall network in training set sample
The branching networks that image is input in overall network are cut, and the loss function that overall network is arranged completes training；
Step 2：Original RGB image input in test set sample is completed to core network in the overall network of training to roll up
Product feature extraction generates multilayer feature figure；
Step 3：Multilayer feature figure is inputted to Area generation network in the overall network for completing training and carries out pedestrian candidate frame
Extraction generates pedestrian candidate region；
Step 4：The classification Recurrent networks completed in the overall network of training are classified and are positioned to pedestrian candidate region
Output includes the detection result image of pedestrian position encirclement frame afterwards.
Preferably, the step 1 includes the following steps：
Step 1.1：The branching networks of core network in overall network initialize and determine loss proportion λi；
Step 1.2：Original RGB image inputs to core network, that corresponding semantic segmentation image is input to branching networks is complete
At foreground sample and background sample selection and generate multilayer feature figure；
Step 1.3：Semantic segmentation loss function L pixel-by-pixel is determined based on the multilayer feature figure of semantic segmentation imagess, used
Formula 1 is as follows：
Wherein, H indicates that characteristic pattern height, W indicate characteristic pattern width, px,yIndicate the feature at characteristic pattern position (x, y)
Value, qx,yIndicate that known corresponding supervisory signals, l (p, q) indicate cross entropy loss function, l (p, q)=- plogq- (1-p)
log(1-q)；
Step 1.4：Multilayer feature figure based on original RGB image carries out the instruction of Area generation network and Recurrent networks of classifying
Practice and determines corresponding loss function；
Step 1.5：Determine that total losses function completes training based on step 1.4 and step 1.3, formula 2 used is as follows：
Wherein, λiIndicate loss proportion,Indicate the Classification Loss function of Area generation network,Indicate region
The candidate frame for generating network positions loss function,The Classification Loss function of presentation class Recurrent networks,It indicates
The positioning loss function of classification Recurrent networks, LSSIndicate semantic segmentation loss function pixel-by-pixel.
Preferably, the step 3 includes the following steps：
Step 3.1：After multilayer feature figure to be inputted to the Area generation network in the overall network for completing training, multilayer feature
Figure size is W*H, and channel is the image of C, and each position based on image generates M*N candidate frame, and wherein W indicates that characteristic pattern is wide
Degree, H indicate that characteristic pattern height, C indicate that image channel number, M indicate that areal array number, N indicate ratio number of combinations；
Step 3.2：Layer of classifying in selection and step 3.1 Area generation network based on foreground sample and background sample exports
Belong to the probability of foreground and background in image on each position in M*N candidate frame, layer belongs to defeated from C dimensional features to classify
Go out the probability i.e. confidence score of encirclement frame of foreground and background；
Step 3.3：Layer, which is returned, based on candidate frame in step 3.1 Area generation network exports M*N candidate on each position
Frame corresponds to window translation zooming parameter, is exported from C dimensional features for the 4 translation contractings of refine candidate frame to which candidate frame returns layer
Put parameter；
Step 3.4：Candidate frame is ranked up according to the confidence score of encirclement frame, selects highest scoring person by non-inhibited
Operation obtains multiple candidate result, that is, pedestrian candidate regions.
Preferably, the step 4 includes the following steps：
Step 4.1：Pedestrian candidate region is inputted the classification Recurrent networks in the overall network for completing training to position and
Classification results complete refine using corresponding 4 translation zooming parameters；
Step 4.2：Finally output includes the detection result image of pedestrian position encirclement frame.
In conclusion by adopting the above-described technical solution, the beneficial effects of the invention are as follows：
1. the present invention, which by the way that loss function is arranged in the training stage, increases, inputs semantic segmentation image progress team surveillance,
It realizes and trained pedestrian's supervisory signals is increased by combination semantic segmentation information under low resolution, the information of more pedestrians is provided,
Help pedestrian to be distinguished from background, it is difficult in low resolution pedestrian and background to solve existing pedestrian detection
The problem for causing accuracy of detection low is distinguished, performance of the pedestrian detection under real scene is improved；
2. the present invention trains whole network, balance semantic segmentation to lose letter by core network and branching networks team surveillance
The loss function accounting of number and core network, provides the supervision message of pixel scale, helps to detach adjacent target, solve
Divided using semantic information based on convolutional neural networks and obscure the problem of adjacent target boundary leads to missing inspection and flase drop, improved close
The shortcomings that being accurately positioned the precision of pedestrian target under crowd's scene of collection, avoiding generating flase drop and missing inspection；
3. the Lss loss functions that the present invention adds are the supervisory signals of pixel scale, finer control information is provided, one
Aspect can improve the setting accuracy of conventional pedestrian, on the other hand be more easily detected the difficult sample that pedestrian's background is not easily distinguishable
This, improves the precision that pedestrian target is accurately positioned under intensive crowd's scene.
Description of the drawings
Examples of the present invention will be described by way of reference to the accompanying drawings, wherein：
Fig. 1 is training and the test block diagram of the present invention；
Fig. 2 is flow chart of the method for the present invention；
Fig. 3 is the input original image of the present invention；
Fig. 4 is the semantic segmentation image of the present invention；
Fig. 5 is the pedestrian detection result output image of the present invention.
Specific implementation mode
All features disclosed in this specification or disclosed all methods or in the process the step of, in addition to mutually exclusive
Feature and/or step other than, can combine in any way.
It elaborates with reference to Fig. 1-5 couples of present invention.
Embodiment 1
A kind of pedestrian detection method based on semantic segmentation information, includes the following steps
Step 1：By core network, the corresponding semanteme point in original RGB image input overall network in training set sample
The branching networks that image is input in overall network are cut, and the loss function that overall network is arranged completes training；
Step 2：Original RGB image input in test set sample is completed to core network in the overall network of training to roll up
Product feature extraction generates multilayer feature figure；
Step 3：Multilayer feature figure is inputted to Area generation network in the overall network for completing training and carries out pedestrian candidate frame
Extraction generates pedestrian candidate region；
Step 4：The classification Recurrent networks completed in the overall network of training are classified and are positioned to pedestrian candidate region
Output includes the detection result image of pedestrian position encirclement frame afterwards.
Embodiment 2
Step 1 includes the following steps：
Step 1.1：The branching networks of core network in overall network initialize and determine loss proportion λi；
Initialization is as follows：The parameter initialization of core network uses pre-training initialization mode, the initialization of branching networks
Using random initializtion, preceding 60,000 iteration, learning rate 0.001, rear 20,000 iteration, using 0.0001, momentum is set learning rate
It is set to 0.9, weight decaying is set as 0.0005, loss proportion λi1 is taken, loss proportion is determined according to different training sets.
Step 1.2：Original RGB image inputs to core network, that corresponding semantic segmentation image is input to branching networks is complete
At foreground sample and background sample selection and generate multilayer feature figure；
Selection foreground sample and background sample are as follows：After demarcating positive negative sample, it is demarcated as positive sample candidate region to each,
It overlaps the maximum candidate frame of ratio and is denoted as foreground sample；The remaining candidate frame of above step, if itself and some calibration weight
Folded ratio is more than 0.7, is denoted as foreground sample；If the overlap proportion of itself and any one calibration is both less than 0.3, it is denoted as background sample
This；Remaining candidate frame is chosen to above step to be discarded；Candidate frame more than image boundary is discarded.
Step 1.3：Semantic segmentation loss function L pixel-by-pixel is determined based on the multilayer feature figure of semantic segmentation imagess, used
Formula 1 is as follows：
Wherein, H indicates that characteristic pattern height, W indicate characteristic pattern width, px,yIndicate the feature at characteristic pattern position (x, y)
Value, qx,yIndicate that known corresponding supervisory signals, l (p, q) indicate cross entropy loss function, l (p, q)=- plogq- (1-p)
log(1-q)；
Step 1.4：Multilayer feature figure based on original RGB image carries out the instruction of Area generation network and Recurrent networks of classifying
Practice and determines corresponding loss function；
Step 1.5：Determine that total losses function completes training based on step 1.4 and step 1.3, formula 2 used is as follows：
Wherein, λiIndicate loss proportion,Indicate the Classification Loss function of Area generation network,Indicate region
The candidate frame for generating network positions loss function,The Classification Loss function of presentation class Recurrent networks,It indicates
The positioning loss function of classification Recurrent networks, LSSIndicate semantic segmentation loss function pixel-by-pixel.
Step 2：Original RGB image input in test set sample is completed to core network in the overall network of training to roll up
Product feature extraction generates multilayer feature figure；
Step 3 includes the following steps：
Step 3.1：After multilayer feature figure to be inputted to the Area generation network in the overall network for completing training, multilayer feature
Figure size is W*H, and channel is the image of C, and each position based on image generates M*N candidate frame, and wherein W indicates that characteristic pattern is wide
Degree, H indicate that characteristic pattern height, C indicate that image channel number, M indicate that areal array number, N indicate ratio number of combinations；The present embodiment M*
N is 3*3, and M combinations include 128*128,256*256,512*512, and N combinations include 1：1、1：2、2：1, according to different training sets
Select numerical value different；
Step 3.2：Layer of classifying in selection and step 3.1 Area generation network based on foreground sample and background sample exports
Belong to the probability of foreground and background in image on each position in M*N candidate frame, layer belongs to defeated from C dimensional features to classify
Go out the probability i.e. confidence score of encirclement frame of foreground and background；
Step 3.3：Layer, which is returned, based on candidate frame in step 3.1 Area generation network exports M*N candidate on each position
Frame corresponds to window translation zooming parameter, is exported from C dimensional features for the 4 translation contractings of refine candidate frame to which candidate frame returns layer
Put parameter；
Step 3.4：Candidate frame is ranked up according to the confidence score of encirclement frame, selects highest scoring person by non-inhibited
Operation obtains multiple candidate result, that is, pedestrian candidate regions.Area generation network is a full convolutional network, and main there are two 1x1
Convolutional layer forms, and the location information of the confidence score of an output encirclement frame, an output encirclement frame refers to coordinate, coordinate i.e. 4
Translate zooming parameter；
Step 4 includes the following steps：
Step 4.1：Pedestrian candidate region is inputted the classification Recurrent networks in the overall network for completing training to position and
Classification results complete refine using corresponding 4 translation zooming parameters；
Step 4.2：Finally output includes the detection result image of pedestrian position encirclement frame.
Effect analysis：According to Fig.5, can be precisely located pedestrian be overlapped blocking position pedestrian, the present invention by
Loss function is arranged in training stage, increases input semantic segmentation image and carries out team surveillance, realizes and pass through joint under low resolution
Semantic segmentation information increases trained pedestrian's supervisory signals, provides the information of more pedestrians, contributes to pedestrian the area from background
It branches away, solve existing pedestrian detection causes accuracy of detection is low to ask in low resolution pedestrian with the differentiation of background difficulty
Topic improves performance of the pedestrian detection under real scene, avoids missing inspection.
Claims (4)
1. a kind of pedestrian detection method based on semantic segmentation information, it is characterised in that：Include the following steps
Step 1：By core network, the corresponding semantic segmentation figure in original RGB image input overall network in training set sample
As the branching networks being input in overall network, and the loss function that overall network is arranged completes training；
Step 2：Original RGB image input in test set sample is completed to core network in the overall network of training and carries out convolution spy
Sign extraction generates multilayer feature figure；
Step 3：Multilayer feature figure is inputted to Area generation network in the overall network for completing training and carries out pedestrian candidate frame extraction
Generate pedestrian candidate region；
Step 4：It completes defeated after the classification Recurrent networks in the overall network of training are classified and positioned to pedestrian candidate region
Go out to include the detection result image of pedestrian position encirclement frame.
2. a kind of pedestrian detection method based on semantic segmentation information according to claim 1, it is characterised in that：The step
Rapid 1 includes the following steps：
Step 1.1：The branching networks of core network in overall network initialize and determine loss proportion λi；
Step 1.2：Before original RGB image input core network, corresponding semantic segmentation image are input to branching networks completion
The selection of scape sample and background sample simultaneously generates multilayer feature figure；
Step 1.3：Semantic segmentation loss function L pixel-by-pixel is determined based on the multilayer feature figure of semantic segmentation imagess, formula used
1 is as follows：
Wherein, H indicates that characteristic pattern height, W indicate characteristic pattern width, px,yIndicate the characteristic value at characteristic pattern position (x, y),
qx,yIndicate that known corresponding supervisory signals, l (p, q) indicate cross entropy loss function, l (p, q)=- plogq- (1-p) log
(1-q)；
Step 1.4：Multilayer feature figure based on original RGB image carries out Area generation network and the training for Recurrent networks of classifying is true
Fixed corresponding loss function；
Step 1.5：Determine that total losses function completes training based on step 1.4 and step 1.3, formula 2 used is as follows：
Wherein, λiIndicate loss proportion,Indicate the Classification Loss function of Area generation network,Indicate Area generation
The candidate frame of network positions loss function,The Classification Loss function of presentation class Recurrent networks,Presentation class
The positioning loss function of Recurrent networks, LSSIndicate semantic segmentation loss function pixel-by-pixel.
3. a kind of pedestrian detection method based on semantic segmentation information according to claim 1, it is characterised in that：The step
Rapid 3 include the following steps：
Step 3.1：After multilayer feature figure to be inputted to the Area generation network in the overall network for completing training, multilayer feature figure is big
Small is W*H, and channel is the image of C, and each position based on image generates M*N candidate frame, and wherein W indicates characteristic pattern width, H
Indicate that characteristic pattern height, C indicate that image channel number, M indicate that areal array number, N indicate ratio number of combinations；
Step 3.2：Layer of classifying in selection and step 3.1 Area generation network based on foreground sample and background sample exports image
In belong to the probability of foreground and background on each position in M*N candidate frame, to classify layer before belonging to output in C dimensional features
The confidence score of probability, that is, encirclement frame of scape and background；
Step 3.3：Layer, which is returned, based on candidate frame in step 3.1 Area generation network exports M*N candidate frame pair on each position
It answers window to translate zooming parameter, is exported from C dimensional features for 4 translation scaling ginsengs of refine candidate frame to which candidate frame returns layer
Number；
Step 3.4：Candidate frame is ranked up according to the confidence score of encirclement frame, highest scoring person is selected to pass through non-inhibited operation
Obtain multiple candidate results i.e. pedestrian candidate region.
4. a kind of pedestrian detection method based on semantic segmentation information according to claim 1, it is characterised in that：The step
Rapid 4 include the following steps：
Step 4.1：Pedestrian candidate region is inputted to the classification Recurrent networks in the overall network for completing training to position and classification
As a result refine is completed using corresponding 4 translation zooming parameters；
Step 4.2：Finally output includes the detection result image of pedestrian position encirclement frame.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810283404.4A CN108446662A (en) | 2018-04-02 | 2018-04-02 | A kind of pedestrian detection method based on semantic segmentation information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810283404.4A CN108446662A (en) | 2018-04-02 | 2018-04-02 | A kind of pedestrian detection method based on semantic segmentation information |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108446662A true CN108446662A (en) | 2018-08-24 |
Family
ID=63198623
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810283404.4A Pending CN108446662A (en) | 2018-04-02 | 2018-04-02 | A kind of pedestrian detection method based on semantic segmentation information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108446662A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109255321A (en) * | 2018-09-03 | 2019-01-22 | 电子科技大学 | A kind of visual pursuit classifier construction method of combination history and instant messages |
| CN109543519A (en) * | 2018-10-15 | 2019-03-29 | 天津大学 | A kind of depth segmentation guidance network for object detection |
| CN109635694A (en) * | 2018-12-03 | 2019-04-16 | 广东工业大学 | A kind of pedestrian detection method, device, equipment and computer readable storage medium |
| CN109784386A (en) * | 2018-12-29 | 2019-05-21 | 天津大学 | A method of it is detected with semantic segmentation helpers |
| CN110008953A (en) * | 2019-03-29 | 2019-07-12 | 华南理工大学 | Potential target Area generation method based on the fusion of convolutional neural networks multilayer feature |
| CN110069986A (en) * | 2019-03-13 | 2019-07-30 | 北京联合大学 | A kind of traffic lights recognition methods and system based on mixed model |
| CN110490058A (en) * | 2019-07-09 | 2019-11-22 | 北京迈格威科技有限公司 | Training method, device, system and the computer-readable medium of pedestrian detection model |
| CN111008613A (en) * | 2019-12-24 | 2020-04-14 | 贺垚凯 | High-density people flow positioning and monitoring method based on field |
| CN111210443A (en) * | 2020-01-03 | 2020-05-29 | 吉林大学 | Deformable convolution mixing task cascading semantic segmentation method based on embedding balance |
| CN111292334A (en) * | 2018-12-10 | 2020-06-16 | 北京地平线机器人技术研发有限公司 | Panoramic image segmentation method and device and electronic equipment |
| CN111340060A (en) * | 2018-12-19 | 2020-06-26 | 财团法人工业技术研究院 | Training method of image generator |
| CN113111732A (en) * | 2021-03-24 | 2021-07-13 | 浙江工业大学 | Method for detecting intensive pedestrians in high-speed service area |
| CN113139549A (en) * | 2021-03-25 | 2021-07-20 | 北京化工大学 | Parameter self-adaptive panorama segmentation method based on multitask learning |
| CN110310256B (en) * | 2019-05-30 | 2021-09-21 | 上海联影智能医疗科技有限公司 | Coronary stenosis detection method, coronary stenosis detection device, computer equipment and storage medium |
| CN114254750A (en) * | 2019-01-29 | 2022-03-29 | 北京金山数字娱乐科技有限公司 | Accuracy loss determination method and apparatus |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106709568A (en) * | 2016-12-16 | 2017-05-24 | 北京工业大学 | RGB-D image object detection and semantic segmentation method based on deep convolution network |
| CN106845430A (en) * | 2017-02-06 | 2017-06-13 | 东华大学 | Pedestrian detection and tracking based on acceleration region convolutional neural networks |
| CN106874894A (en) * | 2017-03-28 | 2017-06-20 | 电子科技大学 | A kind of human body target detection method based on the full convolutional neural networks in region |
| CN107301376A (en) * | 2017-05-26 | 2017-10-27 | 浙江大学 | A kind of pedestrian detection method stimulated based on deep learning multilayer |
| CN107341446A (en) * | 2017-06-07 | 2017-11-10 | 武汉大千信息技术有限公司 | Specific pedestrian's method for tracing and system based on inquiry self-adaptive component combinations of features |
| CN107704866A (en) * | 2017-06-15 | 2018-02-16 | 清华大学 | Multitask Scene Semantics based on new neural network understand model and its application |
-
2018
- 2018-04-02 CN CN201810283404.4A patent/CN108446662A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106709568A (en) * | 2016-12-16 | 2017-05-24 | 北京工业大学 | RGB-D image object detection and semantic segmentation method based on deep convolution network |
| CN106845430A (en) * | 2017-02-06 | 2017-06-13 | 东华大学 | Pedestrian detection and tracking based on acceleration region convolutional neural networks |
| CN106874894A (en) * | 2017-03-28 | 2017-06-20 | 电子科技大学 | A kind of human body target detection method based on the full convolutional neural networks in region |
| CN107301376A (en) * | 2017-05-26 | 2017-10-27 | 浙江大学 | A kind of pedestrian detection method stimulated based on deep learning multilayer |
| CN107341446A (en) * | 2017-06-07 | 2017-11-10 | 武汉大千信息技术有限公司 | Specific pedestrian's method for tracing and system based on inquiry self-adaptive component combinations of features |
| CN107704866A (en) * | 2017-06-15 | 2018-02-16 | 清华大学 | Multitask Scene Semantics based on new neural network understand model and its application |
Non-Patent Citations (3)
| Title |
|---|
| JIAYUAN MAO 等: "What Can Help Pedestrian Detection?", 《2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| S.REN 等: "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| YAOBIN LI 等: "Coarse-to-fine deep neural network for fast pedestrian detection", 《PROCEEDINGS OF SPIE》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109255321B (en) * | 2018-09-03 | 2021-12-10 | 电子科技大学 | Visual tracking classifier construction method combining history and instant information |
| CN109255321A (en) * | 2018-09-03 | 2019-01-22 | 电子科技大学 | A kind of visual pursuit classifier construction method of combination history and instant messages |
| CN109543519A (en) * | 2018-10-15 | 2019-03-29 | 天津大学 | A kind of depth segmentation guidance network for object detection |
| CN109543519B (en) * | 2018-10-15 | 2022-04-15 | 天津大学 | Depth segmentation guide network for object detection |
| CN109635694A (en) * | 2018-12-03 | 2019-04-16 | 广东工业大学 | A kind of pedestrian detection method, device, equipment and computer readable storage medium |
| CN111292334A (en) * | 2018-12-10 | 2020-06-16 | 北京地平线机器人技术研发有限公司 | Panoramic image segmentation method and device and electronic equipment |
| CN111292334B (en) * | 2018-12-10 | 2023-06-09 | 北京地平线机器人技术研发有限公司 | Panoramic image segmentation method and device and electronic equipment |
| CN111340060B (en) * | 2018-12-19 | 2023-03-24 | 财团法人工业技术研究院 | Training method of image generator |
| CN111340060A (en) * | 2018-12-19 | 2020-06-26 | 财团法人工业技术研究院 | Training method of image generator |
| CN109784386B (en) * | 2018-12-29 | 2020-03-17 | 天津大学 | Method for assisting object detection by semantic segmentation |
| CN109784386A (en) * | 2018-12-29 | 2019-05-21 | 天津大学 | A method of it is detected with semantic segmentation helpers |
| CN114254750A (en) * | 2019-01-29 | 2022-03-29 | 北京金山数字娱乐科技有限公司 | Accuracy loss determination method and apparatus |
| CN110069986A (en) * | 2019-03-13 | 2019-07-30 | 北京联合大学 | A kind of traffic lights recognition methods and system based on mixed model |
| CN110008953A (en) * | 2019-03-29 | 2019-07-12 | 华南理工大学 | Potential target Area generation method based on the fusion of convolutional neural networks multilayer feature |
| CN110008953B (en) * | 2019-03-29 | 2023-04-28 | 华南理工大学 | Potential target area generation method based on convolution neural network multi-layer feature fusion |
| CN110310256B (en) * | 2019-05-30 | 2021-09-21 | 上海联影智能医疗科技有限公司 | Coronary stenosis detection method, coronary stenosis detection device, computer equipment and storage medium |
| CN110490058B (en) * | 2019-07-09 | 2022-07-26 | 北京迈格威科技有限公司 | Training method, device and system of pedestrian detection model and computer readable medium |
| CN110490058A (en) * | 2019-07-09 | 2019-11-22 | 北京迈格威科技有限公司 | Training method, device, system and the computer-readable medium of pedestrian detection model |
| CN111008613A (en) * | 2019-12-24 | 2020-04-14 | 贺垚凯 | High-density people flow positioning and monitoring method based on field |
| CN111008613B (en) * | 2019-12-24 | 2023-12-19 | 黑龙江文旅信息科技有限公司 | High-density traffic positioning and monitoring method based on field |
| CN111210443A (en) * | 2020-01-03 | 2020-05-29 | 吉林大学 | Deformable convolution mixing task cascading semantic segmentation method based on embedding balance |
| CN113111732A (en) * | 2021-03-24 | 2021-07-13 | 浙江工业大学 | Method for detecting intensive pedestrians in high-speed service area |
| CN113139549A (en) * | 2021-03-25 | 2021-07-20 | 北京化工大学 | Parameter self-adaptive panorama segmentation method based on multitask learning |
| CN113139549B (en) * | 2021-03-25 | 2024-03-15 | 北京化工大学 | Parameter self-adaptive panoramic segmentation method based on multitask learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108446662A (en) | A kind of pedestrian detection method based on semantic segmentation information | |
| CN106096605B (en) | A kind of image obscuring area detection method and device based on deep learning | |
| CN108986064B (en) | People flow statistical method, equipment and system | |
| CN110533084A (en) | A kind of multiscale target detection method based on from attention mechanism | |
| CN110363134B (en) | Human face shielding area positioning method based on semantic segmentation | |
| WO2020181685A1 (en) | Vehicle-mounted video target detection method based on deep learning | |
| CN106022237B (en) | A kind of pedestrian detection method of convolutional neural networks end to end | |
| CN108062525B (en) | Deep learning hand detection method based on hand region prediction | |
| CN110135296A (en) | Airfield runway FOD detection method based on convolutional neural networks | |
| CN107818302A (en) | Non-rigid multi-scale object detection method based on convolutional neural network | |
| CN106127204A (en) | A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks | |
| CN109993220A (en) | Multi-source Remote Sensing Images Classification method based on two-way attention fused neural network | |
| CN113160062B (en) | Infrared image target detection method, device, equipment and storage medium | |
| CN109584248A (en) | Infrared surface object instance dividing method based on Fusion Features and dense connection network | |
| CN108960404B (en) | Image-based crowd counting method and device | |
| CN106778835A (en) | The airport target by using remote sensing image recognition methods of fusion scene information and depth characteristic | |
| CN105046206B (en) | Based on the pedestrian detection method and device for moving prior information in video | |
| CN109492596B (en) | Pedestrian detection method and system based on K-means clustering and regional recommendation network | |
| CN109543632A (en) | A kind of deep layer network pedestrian detection method based on the guidance of shallow-layer Fusion Features | |
| CN110135513A (en) | A kind of weld joint recognition method of the welding robot based on deep learning | |
| CN110472628A (en) | A kind of improvement Faster R-CNN network detection floating material method based on video features | |
| CN110349167A (en) | A kind of image instance dividing method and device | |
| CN111401293A (en) | Gesture recognition method based on Head lightweight Mask scanning R-CNN | |
| CN112733815B (en) | Traffic light identification method based on RGB outdoor road scene image | |
| CN114170511A (en) | Pavement crack disease identification method based on Cascade RCNN |
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: 20180824 |
|
| RJ01 | Rejection of invention patent application after publication |