CN108304787A - Road target detection method based on convolutional neural networks - Google Patents
Road target detection method based on convolutional neural networks Download PDFInfo
- Publication number
- CN108304787A CN108304787A CN201810044670.1A CN201810044670A CN108304787A CN 108304787 A CN108304787 A CN 108304787A CN 201810044670 A CN201810044670 A CN 201810044670A CN 108304787 A CN108304787 A CN 108304787A
- Authority
- CN
- China
- Prior art keywords
- road target
- target detection
- convolutional neural
- neural networks
- detection method
- 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
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The road target detection method based on convolutional neural networks that the invention discloses a kind of, it is intended to solve that the existing object detection system complicated network structure, accuracy be high, slow-footed technical problem.In the present invention, first make the training set and test set of road target, then TensorFlow deep learning frames are built, establish SSD target detection model structures, feature extraction is carried out to road target image, then loss function optimizing detection model is tested and utilized, finally is classified to obtain testing result using SoftMax algorithms.The beneficial technical effect of the present invention lies in:Network structure is simple, speed is fast and accuracy of detection is high.
Description
Technical field
The present invention relates to target detection technique fields, and in particular to a kind of road target detection based on convolutional neural networks
Method.
Background technology
Road target detection is that road target is found out in road scene image.As auxiliary drives and unpiloted hair
Exhibition, road target detection are also required to have breakthrough as necessary technological means.In driving, vehicle is to surrounding objects object sense
Know that the enhancing of ability can improve the safety of driving.So research road target detection has very important significance.
The research of the target detection of early stage is that researcher's manual extraction target signature is combined, and then utilizes machine
Learning algorithm is detected and identifies.This way depends on the experience of researcher, the quality of the extraction feature of researcher
The effect of detection is directly affected, and these features can have a great impact on Detection accuracy for different graders,
Detection model is set to lack generalization ability.
In recent years, it with the development of deep learning, is examined in image recognition, image based on the deep learning of convolutional neural networks
It surveys and the fields such as image segmentation all achieves prominent achievement.Compared with traditional processing method, the hand to image object is avoided
Work extracts feature, can improve the generalization ability of detection model.Road target detection task be exactly under road complex scene,
Classification and Detection is carried out to the target on image, is then demarcated region respectively.Detection required for being collected under different scenes
Target image data set, from different scenes such as different weather, day and night and environment to the target of different angle and size
Image Acquisition is carried out, ensures the diversity of data set, then by obtaining stable target detection model to the study of data set.
The upsurge of deep learning is lighted in the performance of ImageNet challenge matches from Krizhevsky in 2012 et al., then
Just it has been applied in target detection.The method Selective of extraction object area is utilized in the proposition of R-CNN in 2014
Search and AlexNet finally uses SVM in classification, achieves good effect.Subsequent Kaiming He are proposed can
It solves the problems, such as the too big SPP-net of the amount of computing repeatedly of the convolution feature of R-CNN, has given up Selective Search, made institute
There are district-share convolutional calculation, Ross Girshick in 2015 that will further propose Fast R-CNN, is replaced with softmax recurrence
The expense of room and time is reduced for SVM classifier, Ross Girshick integrate Region Proposal again later
Network (RPN) network, it is proposed that Faster R-CNN, although this method precision increases, speed is not fast enough;
The appearance of YOLO in 2016 realizes end-to-end, multi-task learning, and detection speed is fast, but its precision one to wisp detection
As.
Invention content
There is provided that a kind of design is simple, accuracy is high the technical problem to be solved by the present invention is to overcome the deficiencies in the prior art and
Fireballing road target detection method.
In order to solve the above technical problems, the present invention adopts the following technical scheme that:
A kind of road target detection method based on convolutional neural networks is designed, is included the following steps:
Step 1, the training set and test set for making road target;
Step 2 builds TensorFlow deep learnings frame and introduces VGG network structures, and feature is carried out to road target image
Extraction, the wherein configuration of model are completed using SSD algorithms;
Step 3, in conjunction with ImageNet data sets, multi-target detection problem is converted to road target detection, road target includes
Vehicle, the people of walking and the people that rides;
Step 4 uses cross entropy cost function as confidence level loss function, and the position loss between prediction block and true frame
Function is weighted summation, obtains overall loss function and is optimized to detection model;
Step 5 classifies to obtained characteristic use SoftMax algorithms, the detection model after being trained;
Step 6, by road image input step to be measured(6)In detection model so that obtain testing result.
Further, step 1 includes following processing links:(1)It is following according to the picture statistics in 2007 data sets of VOC
Parameter:Width, height and the ratio of width to height of picture, width, height and the ratio of width to height of road target;(2)The value obtained according to the first step from
Satisfactory picture is filtered out in ImageNet data sets;(3)By the picture making of previous step screening at 2007 data of VOC
Collection.
Further, step 2 includes following processing links:Add various sizes of characteristic pattern, the size of the characteristic pattern
It is gradually reduced, various sizes of effect can be detected to reach.
Further, step 3 includes following processing step:By the num output in program:21 are changed to num
output:4, wherein 4 represent vehicle, 4 class target of pedestrian, the people to ride and background, and corresponding adjustment is made to rest part.
Further, the definition of cross entropy cost function is in step 4:
;WhereinIndicating that i-th of acquiescence frame matches with j-th of true frame that classification is p, p is target category type,
Indicate the confidence level that i-th of acquiescence frame is classification p,。
Further, the definition of smooth loss functions is in step 5:
Wherein:(cx,cy)To give tacit consent to the center of frame, the wide w and high h of frame are given tacit consent to.
Further, overall target loss function is the weighted sum of position loss function and confidence level loss function:
,
Wherein:α=0.5, N are the number for giving tacit consent to frame.
Compared with prior art, the beneficial technical effect of the present invention lies in:
1. present invention design is succinct, build network frame and detection model, after optimized by loss function, structure is apparent,
There is no other cumbersome steps, realizes and be easy.
2. precision of the present invention is high, it is that convolution kernel is used on characteristic pattern that the configuration of detection model, which utilizes SSD algorithms, core,
It predicts classification score, the offset of a series of default bounding boxes, is carried out on the characteristic pattern of different scale
Prediction, realizes end-to-end training, even if can guarantee accuracy of detection if the resolution ratio of image is relatively low.
3. detection speed of the present invention is fast, using the TensorFlow frames of Google, data and model parallelization are good, speed
Soon, and SSD algorithms can directly predict the coordinate and classification of bounding box, eliminate the process for generating proposal, more
Further improve detection speed.
Description of the drawings
Fig. 1 is model learning testing process schematic diagram of the present invention;
Fig. 2 is SSD circuit theory schematic diagrams of the present invention;
Fig. 3 is SSD model trainings process schematic of the present invention;
Fig. 4 is the present invention result figure that road target detects under SSD models;
Fig. 5 is the Precision-Recall curve graphs of present invention training vehicle under SSD models;
Fig. 6 is the Precision-Recall curve graphs of present invention training pedestrian under SSD models;
Fig. 7 is the Precision-Recall curve graphs of the present invention people that training is ridden under SSD models;
Fig. 8 is the statistical information of training set and test set of the present invention;
Fig. 9 is the analysis of experimental results that the present invention is tested under SSD models.
Specific implementation mode
Illustrate the specific implementation mode of the present invention with reference to the accompanying drawings and examples, but following embodiment is used only in detail
It describes the bright present invention in detail, does not limit the scope of the invention in any way.
Involved method or step is unless otherwise instructed then the routine side of the art in following embodiment
Method or step, those skilled in the art can make conventional selection or are adaptively adjusted according to concrete application scene.
The devices such as involved unit module, parts, structure or sensor in following embodiment, unless otherwise instructed,
It is then conventional commercial product.
Embodiment 1:A kind of road target detection method based on convolutional neural networks first makes referring to Fig. 1 to Fig. 3
Then the training set and test set of road target build TensorFlow deep learning frames, establish SSD target detection model knots
Structure carries out feature extraction to road target image, and last test simultaneously utilizes loss function optimizing detection model.Wherein, training set
Production method with test set is:It is counted from 2007 data sets of VOC first:The width of picture, height, the ratio of width to height, road target
Secondly picture in ImageNet data sets is screened according to above-mentioned statistical value, will finally be filtered out by wide, high, the ratio of width to height
The picture making come is at 2007 data sets of VOC, including training set and test set.
It is illustrated in figure 2 the present embodiment SSD algorithm structure schematic diagrames.SSD is to be based on a propagated forward convolutional Neural net
Network generates a series of bounding box of fixed sizes and possibility that each frame includes object example.Later, one is carried out
Non-maxima suppression obtains final prediction.For one with the characteristic layer that p-channel size is m*n, the convolution of 3*3*p is used
Core carries out convolution operation, obtains offset of the acquiescence frame to the confidence level of each target category and this acquiescence frame.With more
A convolution kernel to the characteristic pattern of m*n carry out multiple convolution obtain this characteristic pattern in the confidence level of each classifications of different acquiescence frames and
Coordinate shift value.Acquiescence frame indicates area size of the target representated by artwork herein.In the present embodiment, it increases pair
The extraction of the characteristic pattern of conv3 layers in VGG networks, to enhance the detectability of Small object.Finally press down again with non-maximization
System is screened to obtain final testing result to acquiescence frame.
It is illustrated in figure 3 the present embodiment SSD model training process schematics.The training of model is divided into 2 stages:Pre-training
Stage and training set training stage.The pre-training stage carries out pre-training with ImageNet data sets to the convolutional layer of VGG first, Gu
Determine the convolutional layer for belonging to VGG in SSD models, rest network is trained using VOC2007 data sets, finally to entire model
Network is finely adjusted.In the training set training stage, initialization network is carried out first with pre-training stage obtained parameter, so
Fixed network convolutional layer afterwards is trained the parameter of detection classification, is then finely adjusted to whole network, finally obtains mesh
Mark model.Next test set can be tested with trained object module, obtains the detection result of test set.In order to
Target detection problems in scene are converted into road target by precision and the accuracy for further increasing detection(Vehicle, Hang Renhe
The people to ride)Detection, trained strategy use fine-tuning technologies.
It is the present embodiment result figure that road target detects under SSD models as shown in Figure 4.As shown in Figure 4, for target
Completely, the few object recognition rate of background interference is 0.9 or more, and rainy day, light is weak, target and background is similar and target
Smaller discrimination is relatively low.
It is illustrated in figure 5 the Precision-Recall curve graphs of the present embodiment training vehicle under SSD models.Such as Fig. 6
It show the Precision-Recall curve graphs of the present embodiment training pedestrian under SSD models.It is illustrated in figure 7 the present embodiment
The Precision-Recall curve graphs for the people that training is ridden under SSD models.As seen from the figure, the mAP values for detecting vehicle target are
0.853, the mAP values for detecting pedestrian target are 0.397, and the mAP values for detecting the people's target ridden are 0.682.Analysis learns, vehicle
The effect of detection is preferable, and the people and pedestrian detection effect to ride is general, but can also realize the function of identification.This is because image comes
From the video image recorded in automobile data recorder, the people and pedestrian that ride on highway are walked by roadside, distance recorder
Distance farther out, and pedestrian is farther apart from vehicle relative to the people to ride, so what pedestrian and the people to ride were presented on the image
Target very little, target it is smaller it is easier interfered by background, clarification of objective loses more, the other equipment of automobile data recorder
Quality is general, and most of is the environment and target for focusing on vehicle periphery, poor to remote object shooting quality, at night
Shooting effect in scape is worse, increases the difficulty of identification.
It is illustrated in figure 8 the statistical information of the present embodiment training set and test set.Picture used includes mainly vehicle, pedestrian
With the people to ride.Wherein, the training set of vehicle has 30376 pictures, 14981 pictures of test set;The training set of pedestrian has
1351 pictures, 689 pictures of test set;1026 pictures of people's training set ridden, 596 pictures of test set.That is training set
6263 pictures in total, test set 3333 pictures in total.
It is illustrated in figure 9 the analysis of experimental results tested under the present embodiment SSD models.Increase the size of original image,
Size is set to double, the mAP values of target detection increase the detection that 0.033, SSD models increase the characteristic pattern of conv3, target
The mAP values of detection increase 0.049, and amplification is larger, illustrate to increase picture size and increase conv3 convolution kernels to improve detection essence
Degree, detectability of the enhancing for Small object.
The present invention is described in detail above in conjunction with drawings and examples, still, those of skill in the art
Member is it is understood that without departing from the purpose of the present invention, can also carry out each design parameter in above-described embodiment
Change, forms multiple specific embodiments, is the common variation range of the present invention, is no longer described in detail one by one herein.
Claims (7)
1. a kind of road target detection method based on convolutional neural networks, which is characterized in that include the following steps:
(1)Make the training set and test set of road target;
(2)It builds TensorFlow deep learnings frame and introduces VGG network structures, carrying out feature to road target image carries
It takes, the wherein configuration of detection model is completed using SSD algorithms;
(3)Multi-target detection problem be converted to road target detection, the road target include vehicle, walking people and ride
People;
(4)Use cross entropy cost function as confidence level loss function, and letter is lost in the position between prediction block and true frame
Number is weighted summation, obtains overall loss function and is optimized to detection model;
(5)Classify to obtained characteristic use SoftMax algorithms, obtains testing result.
2. the road target detection method according to claim 1 based on convolutional neural networks, which is characterized in that described
Step(1)In, the making of the training set and test set includes the following steps:
1. counting following parameter according to the picture in 2007 data sets of VOC:Width, height and the ratio of width to height of picture, road target
Wide, high and the ratio of width to height;
2. the value obtained according to upper step filters out satisfactory picture from ImageNet data sets;
3. by the picture making of previous step screening at 2007 data sets of VOC, including training set and test set.
3. the road target detection method according to claim 1 based on convolutional neural networks, which is characterized in that described
Step(2)In, including following processing step:
Various sizes of characteristic pattern is added, the size of the characteristic pattern is gradually reduced, and various sizes of effect can be detected to reach.
4. the road target detection method according to claim 1 based on convolutional neural networks, which is characterized in that described
Step(3)In, including following processing step:
The parameter of num output in detection model is set as 4, wherein 4 represent vehicle, 4 class of pedestrian, the people to ride and background
Target.
5. the road target detection method according to claim 1 based on convolutional neural networks, which is characterized in that described
Step(4)In, the definition of cross entropy cost function is:
;
Wherein,Indicate that i-th of acquiescence frame matches with j-th of true frame that classification is p;P is target category type;Indicate the confidence level that i-th of acquiescence frame is classification p,。
6. the road target detection method according to claim 1 based on convolutional neural networks, which is characterized in that described
Step(4)In, the definition of position loss function is:
Wherein, (cx, cy) is the center for giving tacit consent to frame, gives tacit consent to the width (w) and height (h) of frame.
7. the road target detection method according to claim 1 based on convolutional neural networks, which is characterized in that described
Step(4)In, the function expression of overall loss function is:
,
Wherein:α=0.5, N are the number for giving tacit consent to frame.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810044670.1A CN108304787A (en) | 2018-01-17 | 2018-01-17 | Road target detection method based on convolutional neural networks |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810044670.1A CN108304787A (en) | 2018-01-17 | 2018-01-17 | Road target detection method based on convolutional neural networks |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108304787A true CN108304787A (en) | 2018-07-20 |
Family
ID=62865709
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810044670.1A Pending CN108304787A (en) | 2018-01-17 | 2018-01-17 | Road target detection method based on convolutional neural networks |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108304787A (en) |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109284760A (en) * | 2018-08-10 | 2019-01-29 | 杭州群核信息技术有限公司 | A kind of furniture detection method and device based on depth convolutional neural networks |
CN109344691A (en) * | 2018-08-10 | 2019-02-15 | 广东工业大学 | A kind of visual security campus administration monitoring method based on deep learning |
CN109376738A (en) * | 2018-09-20 | 2019-02-22 | 上海眼控科技股份有限公司 | Altering detecting method based on vehicle VIN code start-stop symbol type |
CN109447018A (en) * | 2018-11-08 | 2019-03-08 | 天津理工大学 | A kind of road environment visual perception method based on improvement Faster R-CNN |
CN109598290A (en) * | 2018-11-22 | 2019-04-09 | 上海交通大学 | A kind of image small target detecting method combined based on hierarchical detection |
CN109614885A (en) * | 2018-11-21 | 2019-04-12 | 齐鲁工业大学 | A kind of EEG signals Fast Classification recognition methods based on LSTM |
CN109657577A (en) * | 2018-12-07 | 2019-04-19 | 重庆理工大学 | A kind of animal detection method based on entropy and motion excursion amount |
CN109741318A (en) * | 2018-12-30 | 2019-05-10 | 北京工业大学 | The real-time detection method of single phase multiple dimensioned specific objective based on effective receptive field |
CN109766769A (en) * | 2018-12-18 | 2019-05-17 | 四川大学 | A kind of road target detection recognition method based on monocular vision and deep learning |
CN109902600A (en) * | 2019-02-01 | 2019-06-18 | 清华大学 | A kind of road area detection method |
CN110166744A (en) * | 2019-04-28 | 2019-08-23 | 南京师范大学 | A kind of monitoring method violating the regulations of setting up a stall based on video geography fence |
CN110287770A (en) * | 2019-05-10 | 2019-09-27 | 中国船舶重工集团公司第七一五研究所 | Individual goal matching and recognition method in a kind of water based on convolutional neural networks |
CN110472503A (en) * | 2019-07-11 | 2019-11-19 | 桂林电子科技大学 | A kind of road target detection method, device and storage medium |
CN110490073A (en) * | 2019-07-15 | 2019-11-22 | 浙江省北大信息技术高等研究院 | Object detection method, device, equipment and storage medium |
CN110728223A (en) * | 2019-10-08 | 2020-01-24 | 济南东朔微电子有限公司 | Helmet wearing identification method based on deep learning |
CN110826514A (en) * | 2019-11-13 | 2020-02-21 | 国网青海省电力公司海东供电公司 | Construction site violation intelligent identification method based on deep learning |
CN110826379A (en) * | 2018-08-13 | 2020-02-21 | 中国科学院长春光学精密机械与物理研究所 | Target detection method based on feature multiplexing and YOLOv3 |
CN111079540A (en) * | 2019-11-19 | 2020-04-28 | 北航航空航天产业研究院丹阳有限公司 | Target characteristic-based layered reconfigurable vehicle-mounted video target detection method |
CN111091110A (en) * | 2019-12-24 | 2020-05-01 | 山东仁功智能科技有限公司 | Wearing identification method of reflective vest based on artificial intelligence |
CN111382787A (en) * | 2020-03-06 | 2020-07-07 | 芯薇(上海)智能科技有限公司 | Target detection method based on deep learning |
CN111444821A (en) * | 2020-03-24 | 2020-07-24 | 西北工业大学 | Automatic identification method for urban road signs |
CN111476129A (en) * | 2020-03-27 | 2020-07-31 | 潍坊申海科技有限公司 | Soil impurity detection method based on deep learning |
CN111639546A (en) * | 2020-05-07 | 2020-09-08 | 金钱猫科技股份有限公司 | Small-scale target cloud computing identification method and device based on neural network |
CN112668696A (en) * | 2020-12-25 | 2021-04-16 | 杭州中科先进技术研究院有限公司 | Unmanned aerial vehicle power grid inspection method and system based on embedded deep learning |
CN113989763A (en) * | 2021-12-30 | 2022-01-28 | 江西省云眼大视界科技有限公司 | Video structured analysis method and analysis system |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106611162A (en) * | 2016-12-20 | 2017-05-03 | 西安电子科技大学 | Method for real-time detection of road vehicle based on deep learning SSD frame |
CN107423760A (en) * | 2017-07-21 | 2017-12-01 | 西安电子科技大学 | Based on pre-segmentation and the deep learning object detection method returned |
CN107578091A (en) * | 2017-08-30 | 2018-01-12 | 电子科技大学 | A kind of Pedestrians and vehicles real-time detection method based on lightweight depth network |
-
2018
- 2018-01-17 CN CN201810044670.1A patent/CN108304787A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106611162A (en) * | 2016-12-20 | 2017-05-03 | 西安电子科技大学 | Method for real-time detection of road vehicle based on deep learning SSD frame |
CN107423760A (en) * | 2017-07-21 | 2017-12-01 | 西安电子科技大学 | Based on pre-segmentation and the deep learning object detection method returned |
CN107578091A (en) * | 2017-08-30 | 2018-01-12 | 电子科技大学 | A kind of Pedestrians and vehicles real-time detection method based on lightweight depth network |
Non-Patent Citations (4)
Title |
---|
SHUZFAN: "基于TensorFlow的SSD车辆检测-1", 《HTTPS://BLOG.CSDN.NET/SHUZFAN/ARTICLE/DETAILS/79034555》 * |
国务院学位委员会办公室,中国科协组织人事部: "《第六届博士生学术年会论文集 上》", 28 February 2009 * |
康圣: "利用多次差影融合法对弱点目标检测和轨迹跟踪", 《激光与红外》 * |
李智勇: "《动态图像分析》", 31 August 1999 * |
Cited By (33)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109344691A (en) * | 2018-08-10 | 2019-02-15 | 广东工业大学 | A kind of visual security campus administration monitoring method based on deep learning |
CN109284760A (en) * | 2018-08-10 | 2019-01-29 | 杭州群核信息技术有限公司 | A kind of furniture detection method and device based on depth convolutional neural networks |
CN110826379A (en) * | 2018-08-13 | 2020-02-21 | 中国科学院长春光学精密机械与物理研究所 | Target detection method based on feature multiplexing and YOLOv3 |
CN110826379B (en) * | 2018-08-13 | 2022-03-22 | 中国科学院长春光学精密机械与物理研究所 | Target detection method based on feature multiplexing and YOLOv3 |
CN109376738A (en) * | 2018-09-20 | 2019-02-22 | 上海眼控科技股份有限公司 | Altering detecting method based on vehicle VIN code start-stop symbol type |
CN109447018B (en) * | 2018-11-08 | 2021-08-03 | 天津理工大学 | Road environment visual perception method based on improved Faster R-CNN |
CN109447018A (en) * | 2018-11-08 | 2019-03-08 | 天津理工大学 | A kind of road environment visual perception method based on improvement Faster R-CNN |
CN109614885A (en) * | 2018-11-21 | 2019-04-12 | 齐鲁工业大学 | A kind of EEG signals Fast Classification recognition methods based on LSTM |
CN109598290A (en) * | 2018-11-22 | 2019-04-09 | 上海交通大学 | A kind of image small target detecting method combined based on hierarchical detection |
CN109657577A (en) * | 2018-12-07 | 2019-04-19 | 重庆理工大学 | A kind of animal detection method based on entropy and motion excursion amount |
CN109657577B (en) * | 2018-12-07 | 2022-06-28 | 重庆理工大学 | Animal detection method based on entropy and motion offset |
CN109766769A (en) * | 2018-12-18 | 2019-05-17 | 四川大学 | A kind of road target detection recognition method based on monocular vision and deep learning |
CN109741318A (en) * | 2018-12-30 | 2019-05-10 | 北京工业大学 | The real-time detection method of single phase multiple dimensioned specific objective based on effective receptive field |
CN109741318B (en) * | 2018-12-30 | 2022-03-29 | 北京工业大学 | Real-time detection method of single-stage multi-scale specific target based on effective receptive field |
CN109902600A (en) * | 2019-02-01 | 2019-06-18 | 清华大学 | A kind of road area detection method |
CN110166744A (en) * | 2019-04-28 | 2019-08-23 | 南京师范大学 | A kind of monitoring method violating the regulations of setting up a stall based on video geography fence |
CN110287770A (en) * | 2019-05-10 | 2019-09-27 | 中国船舶重工集团公司第七一五研究所 | Individual goal matching and recognition method in a kind of water based on convolutional neural networks |
CN110472503A (en) * | 2019-07-11 | 2019-11-19 | 桂林电子科技大学 | A kind of road target detection method, device and storage medium |
CN110490073A (en) * | 2019-07-15 | 2019-11-22 | 浙江省北大信息技术高等研究院 | Object detection method, device, equipment and storage medium |
CN110728223A (en) * | 2019-10-08 | 2020-01-24 | 济南东朔微电子有限公司 | Helmet wearing identification method based on deep learning |
CN110826514A (en) * | 2019-11-13 | 2020-02-21 | 国网青海省电力公司海东供电公司 | Construction site violation intelligent identification method based on deep learning |
CN111079540B (en) * | 2019-11-19 | 2024-03-19 | 北航航空航天产业研究院丹阳有限公司 | Hierarchical reconfigurable vehicle-mounted video target detection method based on target characteristics |
CN111079540A (en) * | 2019-11-19 | 2020-04-28 | 北航航空航天产业研究院丹阳有限公司 | Target characteristic-based layered reconfigurable vehicle-mounted video target detection method |
CN111091110B (en) * | 2019-12-24 | 2023-11-17 | 山东仁功智能科技有限公司 | Reflection vest wearing recognition method based on artificial intelligence |
CN111091110A (en) * | 2019-12-24 | 2020-05-01 | 山东仁功智能科技有限公司 | Wearing identification method of reflective vest based on artificial intelligence |
CN111382787A (en) * | 2020-03-06 | 2020-07-07 | 芯薇(上海)智能科技有限公司 | Target detection method based on deep learning |
CN111444821B (en) * | 2020-03-24 | 2022-03-25 | 西北工业大学 | Automatic identification method for urban road signs |
CN111444821A (en) * | 2020-03-24 | 2020-07-24 | 西北工业大学 | Automatic identification method for urban road signs |
CN111476129A (en) * | 2020-03-27 | 2020-07-31 | 潍坊申海科技有限公司 | Soil impurity detection method based on deep learning |
CN111639546A (en) * | 2020-05-07 | 2020-09-08 | 金钱猫科技股份有限公司 | Small-scale target cloud computing identification method and device based on neural network |
CN112668696A (en) * | 2020-12-25 | 2021-04-16 | 杭州中科先进技术研究院有限公司 | Unmanned aerial vehicle power grid inspection method and system based on embedded deep learning |
CN113989763A (en) * | 2021-12-30 | 2022-01-28 | 江西省云眼大视界科技有限公司 | Video structured analysis method and analysis system |
CN113989763B (en) * | 2021-12-30 | 2022-04-15 | 江西省云眼大视界科技有限公司 | Video structured analysis method and analysis system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108304787A (en) | Road target detection method based on convolutional neural networks | |
CN105512640B (en) | A kind of people flow rate statistical method based on video sequence | |
CN108830188B (en) | Vehicle detection method based on deep learning | |
CN106875373B (en) | Mobile phone screen MURA defect detection method based on convolutional neural network pruning algorithm | |
CN109977812A (en) | A kind of Vehicular video object detection method based on deep learning | |
CN107247956B (en) | Rapid target detection method based on grid judgment | |
CN104537647B (en) | A kind of object detection method and device | |
CN107122776A (en) | A kind of road traffic sign detection and recognition methods based on convolutional neural networks | |
CN108009509A (en) | Vehicle target detection method | |
CN108564030A (en) | Classifier training method and apparatus towards vehicle-mounted thermal imaging pedestrian detection | |
CN113486764B (en) | Pothole detection method based on improved YOLOv3 | |
CN102169543B (en) | People counting method based on multiple source information fusion and on-line learning | |
CN107239762A (en) | Patronage statistical method in a kind of bus of view-based access control model | |
CN110532961B (en) | Semantic traffic light detection method based on multi-scale attention mechanism network model | |
CN106128121B (en) | Vehicle queue length fast algorithm of detecting based on Local Features Analysis | |
CN107133974A (en) | The vehicle type classification method that Gaussian Background modeling is combined with Recognition with Recurrent Neural Network | |
CN104217214A (en) | Configurable convolutional neural network based red green blue-distance (RGB-D) figure behavior identification method | |
CN109492609B (en) | Method for detecting lane line, vehicle and computing equipment | |
CN108764298A (en) | Electric power image-context based on single classifier influences recognition methods | |
CN107977683A (en) | Joint SAR target identification methods based on convolution feature extraction and machine learning | |
CN112270381B (en) | People flow detection method based on deep learning | |
CN109492596A (en) | A kind of pedestrian detection method and system based on K-means cluster and region recommendation network | |
CN111368660A (en) | Single-stage semi-supervised image human body target detection method | |
CN107590427A (en) | Monitor video accident detection method based on space-time interest points noise reduction | |
CN106570490A (en) | Pedestrian real-time tracking method based on fast clustering |
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: 20180720 |
|
RJ01 | Rejection of invention patent application after publication |