CN109543617A - The detection method of intelligent vehicle movement traffic information based on YOLO target detection technique - Google Patents
The detection method of intelligent vehicle movement traffic information based on YOLO target detection technique Download PDFInfo
- Publication number
- CN109543617A CN109543617A CN201811408503.7A CN201811408503A CN109543617A CN 109543617 A CN109543617 A CN 109543617A CN 201811408503 A CN201811408503 A CN 201811408503A CN 109543617 A CN109543617 A CN 109543617A
- Authority
- CN
- China
- Prior art keywords
- training
- intelligent vehicle
- target detection
- traffic information
- convolutional layer
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The detection method of intelligent vehicle movement traffic information based on YOLO target detection technique, it belongs to intelligent vehicle movement traffic information detection field.The present invention solves the problems, such as that the accuracy rate of the method for traditional acquisition intelligent vehicle movement traffic information is lower, slow.Camera by being fixed on certain height by the present invention, the traffic information where intelligent vehicle is acquired at an angle, being fixed on intelligent vehicle with heart-shaped and star indicates intelligent vehicle, barrier indicated by an arrow, indicate terminal with rectangle and triangle, by loading trained model and inputting the collected information of camera to it, can real-time detection go out the traffic information where intelligent vehicle, frame number can reach 40 frame per second, accuracy rate to 95% or more.Because having used deep learning, have robustness high in this way, accuracy rate is high, fireballing advantage.Present invention could apply to intelligent vehicle movement traffic information detection fields to use.
Description
Technical field
The invention belongs to intelligent vehicles to move traffic information detection field, and in particular to a kind of intelligent vehicle movement traffic information
Detection method.
Background technique
With the continuous development and progress of technology, the application of intelligent vehicle also becomes increasingly prevalent, and intelligent vehicle is constantly answered
Used in while bringing convenient to people's lives, new challenge is also brought to the research work of people.Due to intelligent vehicle
Operational process in complete advance, reversing, turn to etc. some basic operations, so in order to guarantee intelligent vehicle driving peace
Entirely, it has to be possible to grasp and understand quickly and in real time the movement traffic information of intelligent vehicle, traditional intelligent vehicle control system
System is often based on front camera acquisition traffic information and then carries out the control of intelligent vehicle movement, but this acquisition intelligence
The accuracy rate of the method for vehicle movement traffic information is lower, speed is slower.
Summary of the invention
The purpose of the present invention is the accuracy rate to solve the method that traditional acquisition intelligent vehicle moves traffic information is lower,
Slow problem.
The technical solution adopted by the present invention to solve the above technical problem is:
The detection method of intelligent vehicle movement traffic information based on YOLO target detection technique, this method includes following step
It is rapid:
Step 1: doing pre-training to YOLO target detection model using ImageNet database, retain after the completion of pre-training
The convolutional layer of YOLO target detection model and maximum pond layer, successively add one averagely after the last one convolutional layer of reservation
Pond layer and a full articulamentum are used for classification task；
It is loaded into the good parameter value of pre-training, and removes the average pond layer and full articulamentum of addition, in the last one convolution
Four convolutional layers and two full articulamentums are successively added after layer again, the weighted value and bias of full articulamentum is initialized, obtains
The good YOLO target detection model of pre-training；
Step 2: acquiring intelligent vehicle traffic information image using camera, every image includes intelligent vehicle, obstacle
Object and terminal tertiary target；
Step 3: to the figure of step 2 acquisition by the way of adding random salt-pepper noise, addition Gaussian noise and translation
As carrying out image data augmentation, the image data after augmentation is divided into training set and test set two parts；
Step 4: training set sample is input in the good YOLO target detection model of pre-training, using computer graphical
Processor GPU training pre-training good YOLO target detection model exports the penalty values of loss function in training process；
With TensorBoard real-time visual current penalty values and accuracy, training loads trained model after stopping,
It is tested with test set, using training result deposit hard disc of computer as final training pattern after the completion of test；
Step 5: the image using camera acquisition comprising intelligent vehicle movement traffic information, defeated by acquired image
Enter and carries out target detection into final training pattern.
The beneficial effects of the present invention are: the present invention provides a kind of, the intelligent vehicle based on YOLO target detection technique is moved
The detection method of traffic information, the present invention acquire intelligent vehicle by the way that camera to be fixed on to certain height at an angle
Traffic information where, being fixed on intelligent vehicle with heart-shaped and star indicates intelligent vehicle, barrier indicated by an arrow, uses
Rectangle and triangle indicate terminal, can by loading trained model and inputting the collected information of camera to it
Real-time detection goes out the traffic information where intelligent vehicle, and frame number can reach 40 frame per second, accuracy rate to 95% or more.Because of fortune
With deep learning, have robustness high in this way, accuracy rate is high, fireballing advantage.
Detailed description of the invention
Fig. 1 is the flow chart of the method for the present invention；
Fig. 2 is the other schematic diagram of picture category used in present invention training；
Fig. 3 is the change curve of total losses value in training process of the present invention；
Fig. 4 is object detection results display diagram of the invention.
Specific embodiment
Specific embodiment 1: embodiment is described with reference to Fig. 1.Based on YOLO target detection described in present embodiment
Technology intelligent vehicle movement traffic information detection method, this method specifically includes the following steps:
Step 1: doing pre-training to YOLO target detection model using ImageNet database, retain after the completion of pre-training
The convolutional layer of YOLO target detection model and maximum pond layer, successively add one averagely after the last one convolutional layer of reservation
Pond layer and a full articulamentum are used for classification task；
It is loaded into the good parameter value of pre-training, and removes the average redization layer and full articulamentum of addition, in the last one convolution
Four convolutional layers and two full articulamentums are successively added after layer again, the weighted value and bias of full articulamentum is initialized, obtains
The good YOLO target detection model of pre-training；
Step 2: acquiring intelligent vehicle traffic information image using camera, every image includes intelligent vehicle, obstacle
Object and terminal tertiary target；
Step 3: to the figure of step 2 acquisition by the way of adding random salt-pepper noise, addition Gaussian noise and translation
As carrying out image data augmentation, the image data after augmentation is divided into training set and test set two parts；
Step 4: training set sample is input in the good YOLO target detection model of pre-training, using computer graphical
Processor GPU training pre-training good YOLO target detection model exports the penalty values of loss function in training process；
With TensorBoard real-time visual current penalty values and accuracy, training loads trained model after stopping,
It is tested with test set, using training result deposit hard disc of computer as final training pattern after the completion of test；
It include intelligence with 35 ° -55 ° of the angle acquisition in depression angle Step 5: being fixed on camera apart from ground 2-3m height
The image of energy vehicle movement traffic information, acquired image is input in final training pattern and carries out target detection.
YOLO algorithm will test problem and regard regression problem as, using single Neural, utilize the information of whole image
Predict the frame of target, identify the classification of target, realize target detection end to end, YOLO compared to algorithm before just like
Lower advantage:
1) very fast.The process of YOLO is simple, and quickly, real-time detection may be implemented in speed；
2) YOLO is predicted using full figure information.Different from sliding window, region proposals, YOLO is being instructed
Practice, utilize full figure information during prediction.Background block is detected as to Fast R-CNN method fault target, reason is Fast
R-CNN method can not see global image when detecting.Compared to Fast R-CNN, YOLO can drop background forecast error rate
It is at half；
3) YOLO may learn the summary information of target.We are using natural image training YOLO, then using art
Image predicts that YOLO is more much higher than the accuracy rate of other algorithm of target detection.
In recent years, deep learning has become a kind of very popular and very extensive application field technology, is calculating
Machine visual field has rapidity, and the object detection method of high-accuracy is its core technology.Figure based on convolutional neural networks
As classification and target detection technique have very big advantage compared with traditional method in accuracy rate and speed.In scene
Complexity, the environment that light intensity changes greatly, the algorithm of target detection based on deep learning have very strong robustness, Neng Goushi
Complex environment is answered to change.
This invention is to acquire entire traffic information, including intelligent vehicle, barrier using the camera for being placed in eminence
And the information such as terminal, it is compared with the traditional method, the algorithm present invention employs YOLO model as target detection, has real-time
Property and rapidity, can be quickly detected intelligent vehicle in image, the important informations such as barrier and terminal.
Specific embodiment 2: the present embodiment is different from the first embodiment in that: the detailed process of step 1 are as follows:
The YOLO target detection model of pre-training is a convolutional neural networks, it includes 20 convolutional layers, 5 pond layers
With 1 full articulamentum, it is respectively:
1 convolution kernel is having a size of 7*7, the convolutional layer that number is 64；
The maximum pond layer of 1 2*2；
1 convolution kernel is having a size of 3*3, the convolutional layer that number is 192；
The maximum pond layer of 1 2*2；
1 convolution kernel is having a size of 1*1, the convolutional layer that number is 128；
1 convolution kernel is having a size of 3*3, the convolutional layer that number is 256；
1 convolution kernel is having a size of 1*1, the convolutional layer that number is 256；
1 convolution kernel is having a size of 3*3, the convolutional layer that number is 512；
The maximum pond layer of 1 2*2；
4 convolution kernels are having a size of 1*1, the convolutional layer that number is 256；
4 convolution kernels are having a size of 3*3, the convolutional layer that number is 512；
1 convolution kernel is having a size of 1*1, the convolutional layer that number is 512；
1 convolution kernel is having a size of 3*3, the convolutional layer that number is 1024；
The maximum pond layer of 1 2*2；
2 convolution kernels are having a size of 1*1, the convolutional layer that number is 512；
2 convolution kernels are having a size of 3*3, the convolutional layer that number is 1024；
The average pond layer of 1 2*2:
One full articulamentum；
Pre-training is done using preceding 20 layer convolutional layer of the ImageNet database to YOLO target detection model, learning rate is set
For 0.0001, batch_size 256, it is constantly iterated the weighted value of the convolutional layer of training adjustment YOLO target detection model
And bias obtains 20 layers of volume before YOLO target detection prescheme when classification task top5 accuracy rate is not less than 85%
The weight of lamination；
An average pond layer and a full articulamentum are successively added after 20 layers of convolutional layer of YOLO target detection model
For classification task, it is loaded into the good parameter value of pre-training, and removes the average pond layer and full articulamentum of addition, at the last one
Four convolutional layers and two full articulamentums are successively added after convolutional layer again；
Four convolutional layers of addition are as follows:
2 convolution kernels are having a size of 3*3, the convolutional layer that number is 1024；
2 convolution kernels are having a size of 3*3, the convolutional layer that number is 1024；
The weighted value and bias for initializing full articulamentum obtain the good YOLO target detection model of pre-training.
Specific embodiment 3: present embodiment is unlike specific embodiment two: the detailed process of step 2 are as follows:
Intelligent vehicle traffic information picture is acquired using camera, every picture includes intelligent vehicle, barrier and end
Point tertiary target；
The picture of camera acquisition is 3000 to 4000, as shown in Fig. 2, camera carries out hand after having acquired picture
Work labels, wherein adopts barrier indicated by an arrow, it is contemplated that the control of intelligent vehicle it needs to be determined that trolley direction, using star
Shape and heart-shaped picture are fixed on intelligent vehicle, to determine the direction of trolley, determine terminal using rectangle and triangle.
Specific embodiment 4: present embodiment is unlike specific embodiment three: the detailed process of step 3 are as follows:
In view of trained data volume is less, so needing to carry out data augmentation；It is relatively strong in order to have trained result
Robustness, the picture of step 2 acquisition is carried out by the way of adding random salt-pepper noise, addition Gaussian noise and translation
Image data augmentation；
Adding random salt-pepper noise is the black and white for adding 500 points in the random site of each picture using the library Opencv
Noise；It to each picture addition Gaussian noise is realized using Gaussian Blur；Translation is by each picture to x-axis and y
Axis positive direction translation 20%；
The image data collection after augmentation is divided into training set and test set according to the ratio of 9:1, training set is for training mind
Through network, test set is used to terminate how examine final training result after training.
Specific embodiment 5: present embodiment is unlike specific embodiment four: the detailed process of step 4 are as follows:
Training set sample is input in the good YOLO target detection model of pre-training, it is tall and handsome up to 1080ti meter using four pieces
Calculation machine graphics processor GPU training pre-training good YOLO target detection model exports the loss of loss function in training process
Value, loss function are as follows:
Wherein:Expression has object in cell i and j-th of bounding box fallout predictor is to the object predicted in cell i
It is responsible for, λcoordFor object specific gravity factor shared in loss function, λnoobjFor bounding box specific gravity shared in loss function
Coefficient, S indicate that each image is divided into S*S (i.e. S2) cell, B indicates that each cell predicts two bounding boxes, xi,
yi,wi,hiIndicate the center point coordinate value of actual object and the width and height of bounding box,Expression predicts
Object center point coordinate value and bounding box width and height, ciIndicate the classification of actual object,Indicate the object predicted
Classification, pi(c) a possibility that indicating that there are objects,Indicate predict a possibility that there are objects, Classes indicate
The type sum of object, c indicate specific type.
Since object and bounding box specific gravity shared in loss function are different, so λcoord、λnoobjFor the two coefficients
It is to prevent from dissipating to distinguish specific gravity shared by the two, the former is typically provided to 5, and the latter 0.5.
With TensorBoard real-time visual current penalty values and accuracy, as shown in figure 3, working as training set penalty values
Deconditioning when lower than 2.00 loads trained model and is tested with test set, if the object for needing to detect on test set is all
It can be detected and Detection accuracy is not less than 95%, then using training result deposit hard disc of computer as final training mould
Type, the training pattern for otherwise loading the deconditioning moment continue to train；
It, will at this time when the object for needing to detect on test set can be detected and Detection accuracy is not less than 95%
The training result deposit hard disc of computer at quarter is as final training pattern.
Specific embodiment 6: present embodiment is unlike specific embodiment five: the computer graphic in step 4
Shape processor GPU is four pieces tall and handsome up to 1080ti computer graphics processor GPU.
Embodiment
Of the invention carries out the fortune that fast target detection is suitable for intelligent carriage under various environment complicated and changeable using YOLO
The dynamic detection of dynamic traffic information, the present invention is by acquisition pre-training data, as shown in Fig. 2, respectively triangle, heart-shaped, star
Shape, rectangle and arrow indicate trolley, the information such as barrier and terminal, for convenience determine trolley and terminal
Direction determines its direction with the line direction at its center, this is for convenience so having used two graphical representation trolleies and terminal
Later to the control of intelligent vehicle.
The present invention acquires pre-training data first, can in simulating actual conditions by putting the above figure with various positions
The various positions situation that can occur, about acquires the picture of three thousand sheets or so, then carries out data augmentation, the mode of data augmentation
There are addition salt-pepper noise, Gaussian noise, translation changes the modes such as exposure and the saturation degree of HSV space, this is to increase number
According to diversity so that the result of model training have robustness, collected data are divided into training set and test set, instruct
Practicing collection is data set used in trained neural network, and test set is training set and test in order to how test final result
The ratio integrated is 9:1.
It loads on pre-training model to YOLO network, initializes full articulamentum, input training data starts to train, use
The penalty values of TensorBoard real-time visual training and the accuracy rate of verifying collection, as shown in figure 3, when penalty values are smaller, and
When accuracy rate is higher, deconditioning loads trained model and is tested with test set data, if result is more satisfied,
Then model is saved, otherwise continues to train.
Trained model is loaded, camera acquired image is inputted, is predicted in real time, as a result as shown in figure 4, place
The frame number of reason can reach 40 frame per second or so, and accuracy rate reaches 95% or more.
The present invention can also have other various embodiments, in the case that is without departing substantially from this Fa Ming Jing Zhong and its essence, this field
Technical staff makes various corresponding changes and modifications in accordance with the present invention, but these corresponding changes and modifications all should belong to
The protection scope of the appended claims of the present invention.
Claims (6)
1. the detection method of the intelligent vehicle movement traffic information based on YOLO target detection technique, which is characterized in that this method
The following steps are included:
Step 1: doing pre-training to YOLO target detection model using ImageNet database, YOLO is retained after the completion of pre-training
The convolutional layer of target detection model and maximum pond layer successively add an average pond after the last one convolutional layer of reservation
Layer and a full articulamentum are used for classification task；
It is loaded into the good parameter value of pre-training, and removes the average pond layer and full articulamentum of addition, after the last one convolutional layer
Four convolutional layers and two full articulamentums are successively added again, are initialized the weighted value and bias of full articulamentum, are obtained pre- instruction
The YOLO target detection model perfected；
Step 2: using camera acquire intelligent vehicle traffic information image, every image include intelligent vehicle, barrier and
Terminal tertiary target；
Step 3: by the way of adding random salt-pepper noise, addition Gaussian noise and translation to the image of step 2 acquisition into
Image data after augmentation is divided into training set and test set two parts by row image data augmentation；
Step 4: training set sample is input in the good YOLO target detection model of pre-training, using computer graphical processing
Device GPU training pre-training good YOLO target detection model exports the penalty values of loss function in training process；
With TensorBoard real-time visual current penalty values and accuracy, training loads trained model after stopping, with survey
Examination collection is tested, using training result deposit hard disc of computer as final training pattern after the completion of test；
Step 5: the image using camera acquisition comprising intelligent vehicle movement traffic information, acquired image is input to
Target detection is carried out in final training pattern.
2. the detection side of the intelligent vehicle movement traffic information according to claim 1 based on YOLO target detection technique
Method, which is characterized in that the detailed process of the step 1 are as follows:
The YOLO target detection model of pre-training is a convolutional neural networks, it includes 20 convolutional layers, 5 pond layers and 1
A full articulamentum is respectively:
1 convolution kernel is having a size of 7*7, the convolutional layer that number is 64；
The maximum pond layer of 1 2*2；
1 convolution kernel is having a size of 3*3, the convolutional layer that number is 192；
The maximum pond layer of 1 2*2；
1 convolution kernel is having a size of 1*1, the convolutional layer that number is 128；
1 convolution kernel is having a size of 3*3, the convolutional layer that number is 256；
1 convolution kernel is having a size of 1*1, the convolutional layer that number is 256；
1 convolution kernel is having a size of 3*3, the convolutional layer that number is 512；
The maximum pond layer of 1 2*2；
4 convolution kernels are having a size of 1*1, the convolutional layer that number is 256；
4 convolution kernels are having a size of 3*3, the convolutional layer that number is 512；
1 convolution kernel is having a size of 1*1, the convolutional layer that number is 512；
1 convolution kernel is having a size of 3*3, the convolutional layer that number is 1024；
The maximum pond layer of 1 2*2；
2 convolution kernels are having a size of 1*1, the convolutional layer that number is 512；
2 convolution kernels are having a size of 3*3, the convolutional layer that number is 1024；
The average pond layer of 1 2*2:
One full articulamentum；
Pre-training is done using preceding 20 layer convolutional layer of the ImageNet database to YOLO target detection model, setting learning rate is
0.0001, batch_size 256, be constantly iterated training adjustment YOLO target detection model convolutional layer weighted value with
And bias obtains 20 layers of convolution before YOLO target detection prescheme when classification task top5 accuracy rate is not less than 85%
The weight of layer；
An average pond layer is successively added after 20 layers of convolutional layer of YOLO target detection model and a full articulamentum is used for
Classification task is loaded into the good parameter value of pre-training, and removes the average pond layer and full articulamentum of addition, in the last one convolution
Four convolutional layers and two full articulamentums are successively added after layer again；
Four convolutional layers of addition are as follows:
2 convolution kernels are having a size of 3*3, the convolutional layer that number is 1024；
2 convolution kernels are having a size of 3*3, the convolutional layer that number is 1024；
The weighted value and bias for initializing full articulamentum obtain the good YOLO target detection model of pre-training.
3. the detection side of the intelligent vehicle movement traffic information according to claim 2 based on YOLO target detection technique
Method, which is characterized in that the detailed process of the step 2 are as follows:
Intelligent vehicle traffic information picture is acquired using camera, every picture includes intelligent vehicle, barrier and terminal three
Class target；
The picture of camera acquisition is 3000 to 4000, and camera label by hand after having acquired picture, wherein
Barrier indicated by an arrow is adopted, is fixed on intelligent vehicle using star and heart-shaped picture, is determined using rectangle and triangle
Terminal.
4. the detection side of the intelligent vehicle movement traffic information according to claim 3 based on YOLO target detection technique
Method, which is characterized in that the detailed process of the step 3 are as follows:
Picture number is carried out to the picture of step 2 acquisition by the way of adding random salt-pepper noise, addition Gaussian noise and translation
According to augmentation；
Adding random salt-pepper noise is to be made an uproar using the library Opencv in the black and white that the random site of each picture adds 500 points
Sound；It to each picture addition Gaussian noise is realized using Gaussian Blur；Translation is by each picture to x-axis and y-axis
Positive direction translation 20%；
The image data collection after augmentation is divided into training set and test set according to the ratio of 9:1.
5. the detection side of the intelligent vehicle movement traffic information according to claim 4 based on YOLO target detection technique
Method, which is characterized in that the detailed process of the step 4 are as follows:
Training set sample is input in the good YOLO target detection model of pre-training, is instructed using computer graphics processor GPU
Practice the good YOLO target detection model of pre-training, the penalty values of loss function, loss function exported in training process are as follows:
Wherein:Expression has object in cell i and j-th of bounding box fallout predictor is negative to the object predicted in cell i
Duty, λcoordFor object specific gravity factor shared in loss function, λnoobjFor bounding box specific gravity system shared in loss function
Number, S indicate that each image is divided into the cell of S*S, and B indicates that each cell predicts two bounding boxes, xi,yi,wi,hi
Indicate the center point coordinate value of actual object and the width and height of bounding box,Indicate the object predicted
The width and height of center point coordinate value and bounding box, ciIndicate the classification of actual object,Indicate the classification of the object predicted,
pi(c) a possibility that indicating that there are objects,Indicate to predict a possibility that there are objects, Classes indicates object
Type sum, c indicate specific type；
With TensorBoard real-time visual current penalty values and accuracy, stop when training set penalty values are lower than 2.00
Training, loads trained model and is tested with test set, if the object for needing to detect on test set can be detected and examine
It surveys accuracy rate and is not less than 95%, then using training result deposit hard disc of computer as final training pattern, otherwise load stops
The training pattern at training moment continues to train；
When the object for needing to detect on test set can be detected and Detection accuracy is not less than 95%, by this moment
Training result is stored in hard disc of computer as final training pattern.
6. the detection side of the intelligent vehicle movement traffic information according to claim 5 based on YOLO target detection technique
Method, which is characterized in that computer graphics processor GPU in the step 4 is four pieces tall and handsome up at 1080ti computer graphical
Manage device GPU.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811408503.7A CN109543617A (en) | 2018-11-23 | 2018-11-23 | The detection method of intelligent vehicle movement traffic information based on YOLO target detection technique |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811408503.7A CN109543617A (en) | 2018-11-23 | 2018-11-23 | The detection method of intelligent vehicle movement traffic information based on YOLO target detection technique |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109543617A true CN109543617A (en) | 2019-03-29 |
Family
ID=65849836
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811408503.7A Pending CN109543617A (en) | 2018-11-23 | 2018-11-23 | The detection method of intelligent vehicle movement traffic information based on YOLO target detection technique |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109543617A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222557A (en) * | 2019-04-22 | 2019-09-10 | 北京旷视科技有限公司 | Real-time detection method, device, system and the storage medium of road conditions |
| CN110604597A (en) * | 2019-09-09 | 2019-12-24 | 李胜利 | Method for intelligently acquiring fetal cardiac cycle images based on ultrasonic four-cavity cardiac section |
| CN112329768A (en) * | 2020-10-23 | 2021-02-05 | 上善智城(苏州)信息科技有限公司 | Improved YOLO-based method for identifying fuel-discharging stop sign of gas station |
| CN113034497A (en) * | 2021-04-26 | 2021-06-25 | 博科视(苏州)技术有限公司 | Vision-based thermos cup weld positioning detection method and system |
| CN115311869A (en) * | 2022-07-29 | 2022-11-08 | 重庆长安汽车股份有限公司 | Road event obtaining method and system based on intelligent network connection automobile computing power sharing and automobile |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106327920A (en) * | 2016-08-31 | 2017-01-11 | 深圳天珑无线科技有限公司 | Road condition detection method and road condition detection system |
| CN108229366A (en) * | 2017-12-28 | 2018-06-29 | 北京航空航天大学 | Deep learning vehicle-installed obstacle detection method based on radar and fusing image data |
| CN108256554A (en) * | 2017-12-20 | 2018-07-06 | 深圳市金溢科技股份有限公司 | Vehicle reverse stopping judgment method, server and system based on deep learning |
-
2018
- 2018-11-23 CN CN201811408503.7A patent/CN109543617A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106327920A (en) * | 2016-08-31 | 2017-01-11 | 深圳天珑无线科技有限公司 | Road condition detection method and road condition detection system |
| CN108256554A (en) * | 2017-12-20 | 2018-07-06 | 深圳市金溢科技股份有限公司 | Vehicle reverse stopping judgment method, server and system based on deep learning |
| CN108229366A (en) * | 2017-12-28 | 2018-06-29 | 北京航空航天大学 | Deep learning vehicle-installed obstacle detection method based on radar and fusing image data |
Non-Patent Citations (5)
| Title |
|---|
| JOSEPH REDMON等: "You Only Look Once: Unified, Real-Time Object Detection", 《2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
| -牧野-: "YOLO v1检测原理", 《HTTPS://BLOG.CSDN.NET/DCRMG/ARTICLE/DETAILS/81582620》 * |
| 王宇宁等: "基于YOLO算法的车辆实时检测", 《武汉理工大学学报》 * |
| 王超等: "基于YOLO v2模型的交通标识检测算法", 《计算机应用》 * |
| 黎洲等: "基于YOLO＿v2模型的车辆实时检测", 《中国机械工程》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222557A (en) * | 2019-04-22 | 2019-09-10 | 北京旷视科技有限公司 | Real-time detection method, device, system and the storage medium of road conditions |
| CN110222557B (en) * | 2019-04-22 | 2021-09-21 | 北京旷视科技有限公司 | Real-time road condition detection method, device and system and storage medium |
| CN110604597A (en) * | 2019-09-09 | 2019-12-24 | 李胜利 | Method for intelligently acquiring fetal cardiac cycle images based on ultrasonic four-cavity cardiac section |
| CN112329768A (en) * | 2020-10-23 | 2021-02-05 | 上善智城(苏州)信息科技有限公司 | Improved YOLO-based method for identifying fuel-discharging stop sign of gas station |
| CN113034497A (en) * | 2021-04-26 | 2021-06-25 | 博科视(苏州)技术有限公司 | Vision-based thermos cup weld positioning detection method and system |
| CN115311869A (en) * | 2022-07-29 | 2022-11-08 | 重庆长安汽车股份有限公司 | Road event obtaining method and system based on intelligent network connection automobile computing power sharing and automobile |
| CN115311869B (en) * | 2022-07-29 | 2023-07-18 | 重庆长安汽车股份有限公司 | Road event acquisition method and system based on intelligent network-connected automobile calculation force sharing and automobile |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ma et al. | Automatic detection and counting system for pavement cracks based on PCGAN and YOLO-MF | |
| CN109543617A (en) | The detection method of intelligent vehicle movement traffic information based on YOLO target detection technique | |
| CN112070729B (en) | Anchor-free remote sensing image target detection method and system based on scene enhancement | |
| CN108171112A (en) | Vehicle identification and tracking based on convolutional neural networks | |
| CN110533084A (en) | A kind of multiscale target detection method based on from attention mechanism | |
| CN109559302A (en) | Pipe video defect inspection method based on convolutional neural networks | |
| CN107742099A (en) | A kind of crowd density estimation based on full convolutional network, the method for demographics | |
| CN110298387A (en) | Incorporate the deep neural network object detection method of Pixel-level attention mechanism | |
| CN108647588A (en) | Goods categories recognition methods, device, computer equipment and storage medium | |
| CN107480730A (en) | Power equipment identification model construction method and system, the recognition methods of power equipment | |
| CN109558815A (en) | A kind of detection of real time multi-human face and tracking | |
| CN110363104A (en) | A kind of detection method of diesel oil black smoke vehicle | |
| CN107529650A (en) | Network model construction and closed loop detection method, corresponding device and computer equipment | |
| CN106469304A (en) | Handwritten signature location positioning method in bill based on depth convolutional neural networks | |
| CN111242144B (en) | Method and device for detecting abnormality of power grid equipment | |
| Cepni et al. | Vehicle detection using different deep learning algorithms from image sequence | |
| CN113111727B (en) | Feature alignment-based method for detecting rotating target in remote sensing scene | |
| CN112560675B (en) | Bird visual target detection method combining YOLO and rotation-fusion strategy | |
| CN108230354A (en) | Target following, network training method, device, electronic equipment and storage medium | |
| CN109886147A (en) | A kind of more attribute detection methods of vehicle based on the study of single network multiple-task | |
| Liu et al. | Coastline extraction method based on convolutional neural networks—A case study of Jiaozhou Bay in Qingdao, China | |
| CN110390673A (en) | Cigarette automatic testing method based on deep learning under a kind of monitoring scene | |
| CN116152685B (en) | Pedestrian detection method and system based on unmanned aerial vehicle visual field | |
| CN116824335A (en) | YOLOv5 improved algorithm-based fire disaster early warning method and system | |
| CN114511710A (en) | Image target detection method based on convolutional neural network |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 150001 Harbin, Nangang, West District, large straight street, No. 92 Applicant after: Yu Xinghu Address before: 264006 No. 32, Yantai economic and Technological Development Zone, Shandong, Zhujianglu Road Applicant before: Yu Xinghu |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190329 |