CN111488854A - Automatic identification and classification method for road traffic signs - Google Patents
Automatic identification and classification method for road traffic signs Download PDFInfo
- Publication number
- CN111488854A CN111488854A CN202010329229.5A CN202010329229A CN111488854A CN 111488854 A CN111488854 A CN 111488854A CN 202010329229 A CN202010329229 A CN 202010329229A CN 111488854 A CN111488854 A CN 111488854A
- Authority
- CN
- China
- Prior art keywords
- road
- images
- signs
- mask
- image
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—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
- G06V20/582—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of traffic signs
-
- 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
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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)
- 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)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to an automatic identification and classification method for road traffic signs, which comprises the following steps: step S1: collecting road images by adopting vehicle-mounted image collecting equipment; step S2: screening out images with traffic signs from the acquired road images, marking the images, and constructing a data set required by Mask _ RCNN model training; step S3: inputting the data set obtained in the step S2 into a Mask _ RCNN model for training to obtain a trained weight; step S4: carrying out identification and classification on road signs on all the collected road images by using the weights trained in the step S3; step S5: checking the generated result, carrying out secondary annotation on the image with poor recognition effect and retraining the weight; step S6: and outputting the recognition result. The method is beneficial to improving the recognition and classification effects of the road traffic signs.
Description
Technical Field
The invention relates to the technical field of road automatic detection, in particular to an automatic identification and classification method for road traffic signs.
Background
At present, the detection and identification of traffic signs are the research hotspots of scholars and institutions at home and abroad. Generally, there are 3 methods for detecting traffic signs: color segmentation based, shape information based, and machine learning based methods. The algorithm based on color segmentation is simple, the calculation speed is high, and the algorithm is insensitive to geometric deformation, but the defects are obvious in scenes such as low-illumination or backlight environment and similar background, because the color is unreliable information, the colors collected at different time and under different illumination are different. The algorithm based on the shape information is simple and fast, but other objects with the same shape are easily and wrongly recognized as the traffic signs, and the traffic signs often have the same shape corresponding to a plurality of different types, for example, the rectangular traffic signs comprise the indication signs, the direction signs and the notification signs, so that the recognition and the classification of the traffic signs cannot be accurately realized by using the shape information.
In the process of carrying out mark recognition based on machine learning, with the rise of intelligent transportation systems, deep learning begins to be added into a mark recognition method. Convolutional neural networks are widely used in image detection, from the first R-CNN to FastR-CNN to Faster R-CNN and finally to today's Mask R-CNN. The development from the initial single target to the multi-target recognition of the present day has also greatly improved in speed.
Although the road traffic sign types in China are approximately fixed, the road traffic signs in different areas can be more or less different when being designed. In order to avoid the difference of traffic signs due to different regions, the data sets of traffic signs required for different cities are different. Deep learning performs recognition and classification by learning the features of a target object, and is likely to cause erroneous judgment when there is an object having similar features.
Disclosure of Invention
The invention aims to provide an automatic identification and classification method for road traffic signs, which is beneficial to improving the identification and classification effects of the road traffic signs.
In order to achieve the purpose, the invention adopts the technical scheme that: an automatic identification and classification method for road traffic signs comprises the following steps:
step S1: collecting road images by adopting vehicle-mounted image collecting equipment;
step S2: screening out images with traffic signs from the acquired road images, marking the images, and constructing a data set required by Mask _ RCNN model training;
step S3: inputting the data set obtained in the step S2 into a Mask _ RCNN model for training to obtain a trained weight;
step S4: carrying out identification and classification on road signs on all the collected road images by using the weights trained in the step S3;
step S5: checking the generated result, carrying out secondary annotation on the image with poor recognition effect and retraining the weight;
step S6: and outputting the recognition result.
Further, the step S2 specifically includes the following steps:
step S21: determining the traffic sign types: according to a mark information acquisition table of a highway maintenance center in Fujian province, and by combining acquired road images, the traffic mark types of roads are divided into six major categories of a notice mark, a prohibition mark, a warning mark, a road indication mark, an indication mark and a milepost;
and S22, labeling the road images through L abelme labeling software, selecting images with traffic signs from the acquired road images to label the images to obtain labeled images, labeling the traffic signs by using English names during labeling, wherein the English names corresponding to the notification signs, the prohibition signs, the warning signs, the road signs, the indication signs and the mileposts are respectively notice, promotion, warning, guide, indication and milestone, and if a plurality of signs of the same type appear in one road image, adding serial numbers after the names.
Further, in the step S3, the Mask _ RCNN model mainly includes four main parts: (1) the main frame is a standard convolution neural network, the bottom layer detects low-level characteristics, and the high layer detects high-level characteristics; allowing the feature of each level to be combined with the high-level and low-level features during the propagation process of the convolutional neural network; (2) the region suggestion network RPN is a light neural network, scans images through a sliding window and finds a region where a target exists; (3) the ROI classifier and the boundary frame regressor are used for dividing the categories of the target objects on the basis of obtaining the target area based on the RPN; the ROI classifier can classify the region into a specific category and can also generate a background category; further fine-tuning the position and size of the bounding box by using a bounding box regressor to encapsulate the target; (4) dividing masks, wherein a mask branch is a convolution network, taking the positive regions selected by the ROI classifier as input, and generating masks of the positive regions; in the training process, the real mask is reduced to calculate a loss function, in the inference process, the predicted mask is enlarged to the size of an ROI frame to give a final mask result, and each target has one mask;
and (5) transmitting the image marked in the step (S2) into a Mask _ RCNN model for training to obtain a trained weight.
Further, in step S4, the weights obtained by training in step S3 are saved in the file positions corresponding to the collected road images; setting a form of an output image, and if the content of the label is more than that of the traffic sign label set at the beginning, not displaying a mask of the label of the non-traffic sign on the image; and then, all the collected road images are imported into a model for training, and a result image with traffic sign category and position information is automatically generated.
Further, the step S5 specifically includes the following steps:
step S51: checking the result generated in the step S4, and screening out the image with error identification;
step S52: re-labeling the data set for the false detection image appearing in step S51, specifically: returning to the step S2, adding two categories of windows and billboards on the basis of the original marking, and respectively naming the two categories as window and billboard; marking the two newly added classes, and retraining the marked image data; generating a new training result and a new recognition result;
step S53: and checking the new generated result until the recognition effect is in accordance with the expectation.
Further, in step S6, inputting the picture road traffic sign information into an Excel table, the method includes: and importing the road mileage information obtained in the acquisition into an Excel table, importing the identified traffic sign type into the Excel table, and importing the identified traffic sign position information into the Excel table.
Compared with the prior art, the invention has the following beneficial effects: the method identifies the traffic signs through a Mask _ RCNN model, and inspects and optimizes the identification result, so that the method can effectively identify and classify the traffic signs and obtain the position information of the traffic signs on images, thereby providing a certain basis for vehicle-assisted driving and unmanned driving, and having strong practicability and wide application prospect.
Drawings
FIG. 1 is a flow chart of a method implementation of an embodiment of the present invention.
Fig. 2 is a road image acquired in the embodiment of the present invention.
FIG. 3 is a diagram of the Mask _ RCNN model architecture according to an embodiment of the present invention.
Fig. 4 shows the recognition result of the traffic sign image according to the embodiment of the invention.
FIG. 5 is Excel table information in an embodiment of the present invention.
Detailed Description
The invention is described in further detail below with reference to the figures and the embodiments.
The invention provides an automatic identification and classification method for road traffic signs, which comprises the following steps as shown in figure 1:
step S1: and acquiring road images by adopting vehicle-mounted image acquisition equipment (vehicle-mounted camera). Fig. 2 shows a road image captured by a camera.
Step S2: and screening images with traffic signs from the acquired road images, labeling the images, and constructing a data set required by Mask _ RCNN model training. The method specifically comprises the following steps:
step S21: determining traffic sign categories
Traffic signs are roughly divided into seven categories, namely, bulletin signs, forbidden signs, warning signs, tourist areas signs, road signs, indicating signs and operation area signs according to a sign information acquisition table of a highway maintenance center in Fujian province. The actual acquisition road section has no tourism area mark and operation area mark. The road mileage stake can be used for correcting errors occurring in the mileage in the acquisition process, and has a special effect on road sign positioning, so in the embodiment, the mileage stake is used as a traffic sign for marking. In summary, in the embodiment, according to the mark information acquisition table of the highway maintenance center in the Fujian province, the traffic mark types of the roads are divided into six major types, namely, a notification mark, a prohibition mark, a warning mark, a direction mark, an indication mark and a milepost, by combining the acquired road images;
step S22, labeling the road image through L abelme labeling software
And selecting an image with a traffic sign from the acquired road images for labeling to obtain a labeled image. When selecting, a certain number of types should be considered, otherwise, a certain traffic sign class is few, and the generated weight cannot identify the type well. For convenience of subsequent processing, the traffic sign is marked by using an English name during marking, and the English names corresponding to the notice sign, the prohibition sign, the warning sign, the road-indicating sign, the indicating sign and the mileage stake are respectively notice, promotion, warning, guide, indication and milestone. If a plurality of marks of the same type appear in one road image, the serial numbers are added after the names in sequence. And decompressing the image data marked and storing the image data into a specified folder.
A) L abelme annotation software usage
L abelme software is image labeling software of a graphical interface, which is written by Python language, PyQt L abelme can label the image in the forms of polygon, rectangle, circle, multi-segment line, line segment, point, etc. for target detection and image segmentation, and label the image in the form of flag for image classification.
B) Labeling collected samples
And (4) collecting samples for screening. In the acquisition process, due to the discontinuity of the traffic sign setting, the traffic sign does not appear on each road image, and when the traffic sign is labeled, the image with the traffic sign needs to be labeled, and a part of the road image with the traffic sign needs to be screened as a training set. In the screening process, a certain number of each traffic sign is ensured according to the classification of the traffic signs, so that the condition that the trained weight cannot be identified or the result that the traffic identification is poor is avoided.
And (5) labeling of the data set. And labeling the screened images by using labelme software, and adding serial numbers to the names of the marks in sequence if a plurality of marks of the same type appear on the same image. And (5) arranging the marked images and using the images for training the model.
The invention is based on a Mask _ RCNN model, and the model can realize the identification and classification of the target object. And putting the marked images into a model for training to generate trained weights. And traversing the acquired images on the trained weights. The image output by the model can classify the marks on the image and generate a mask at the corresponding position, so that people can conveniently observe the marks. The relevant steps involved are as follows.
Step S3: and (5) inputting the data set obtained in the step (S2) into a Mask _ RCNN model for training to obtain the trained weight.
As shown in FIG. 3, the Mask _ RCNN model mainly comprises four main parts, namely (1) a backbone architecture which is a standard convolutional neural network, low-level features (edges and corners) are detected at the bottom layer, high-level features (cars and sky) are detected at the high layer, each level of features are allowed to be combined with the high-level and low-level features in the convolutional neural network propagation process, (2) a region suggestion network RPN which is a light-weight neural network and scans images through sliding windows to find regions where targets exist, (3) an ROI classifier and a bounding box regressor which divides the classes of the targets on the basis of the target regions obtained based on the RPN, the ROI classifier can classify the regions into specific classes and can generate a background class, the positions and the sizes of borders are further finely adjusted by using the bounding box regressor to encapsulate the targets, (4) Mask segmentation, Mask branches are a convolutional network, the positive regions selected by the ROI classifier are taken as input and generate masks, masks which are low-resolution 28 pixels, but are soft-point masks represented by floating-point numbers, the Mask branches are soft masks, the Mask models which are helpful for calculating the final Mask loss prediction process of the final Mask in a small-based on-scale prediction of the Mask prediction process, and the prediction process is performed by using a small-based on the prediction process of the prediction of the Mask-based on the prediction of the Mask model.
And writing the number of the label classifications and the name of each type in the model, setting related parameters such as training rounds, learning rate and the like, and transmitting the image marked in the step S2 into a Mask _ RCNN model for training to obtain the trained weight. And using the trained weight for the recognition of all road images, and setting a corresponding folder to store the generated result.
Step S4: and (4) carrying out identification and classification on road signs on all the acquired road images by using the weights trained in the step S3. Fig. 4 shows the recognition result of the present invention.
Saving the weight obtained by training in the step S3 to a file position corresponding to the acquired road image; setting a form of an output image, and if the content of the label is more than that of the traffic sign label set at the beginning, not displaying a mask of the label of the non-traffic sign on the image; and then, all the collected road images are imported into a model for training, and a result image with traffic sign category and position information is automatically generated.
Step S5: and checking the generated result, carrying out secondary labeling on the image with poor recognition effect and retraining the weight.
In the deep learning process, the model learns the characteristics of the target and stores the characteristics as weights, so that some targets with different types but similar characteristics are identified as the same type. Since the traffic sign recognition is based on the features of the traffic sign, such as color, shape, etc., when features similar to the traffic sign appear on the image, the model is easily misdetected. In the invention, the road sign in the traffic sign type is mainly based on green, and white and a small amount of blue appear on part of the road sign. Many billboards and windows are similar in color to such traffic signs, and therefore, some of the billboards with green and white colors and windows of surrounding premises are mistakenly detected as road signs during the identification process. In order to avoid the wrong identification and increase the discriminativity between the billboard, the window and the road sign, the invention considers that the billboard and the window are subjected to secondary image labeling and are placed into a model for training, the weight capable of identifying the window and the billboard is generated, and the wrong classification is avoided. In summary, the steps specifically include the following steps:
step S51: the result generated in step S4 is checked to screen out and analyze the image with the error.
The Mask _ RCNN model is used for learning through the traffic sign features marked in the training set, and judging whether the traffic signs exist on the images and the types of the traffic signs according to the learning result. When the characteristics of the interfering object are similar to those of the traffic sign, a case of a wrong judgment may occur. In this example, the partial way sign exhibits a rectangular, white character, which is quite familiar with the character of windows and partial white billboards. Through the analysis of the recognition result, the interferents are a white billboard, a window of a surrounding house, a front license plate number, a logo on the surface of a truck, a red safety helmet of a worker and the like, wherein the white billboard and the window are the most frequent.
Step S52: the data set is relabeled for the false positive image that occurred in step S51.
And aiming at the interferents which are easy to be identified as the direction marks, carrying out secondary labeling on the interferents and carrying out weight training again. The main interferents of the invention are the billboard and the window, in order to make the model not recognize these two types of interferents as the direction mark, the procedure returns to step S2, and on the basis of the original label, two categories of the billboard and the window are newly added in the category, which are respectively expressed by the english names billboard and window. And in the previously marked image, checking whether the image contains the two types one by one, and if so, selecting a corresponding marking method according to the shape of the image to mark. And putting the marked images into the model for secondary training of the weight, and identifying the traffic sign by using the generated new weight. In the output process, if the identification object is a window or a billboard, the information of the identification object is not output, so that the traffic sign with the billboard and the window removed is obtained. And marking the two newly added classes, retraining the marked image data, and generating a new training result and a new recognition result. The method can better improve the accuracy of identification.
Step S53: and checking the new generated result until the recognition effect is in accordance with the expectation.
Step S6: and outputting the recognition result.
In order to know the distribution of the road traffic signs exactly and facilitate the subsequent statistics and processing, as shown in fig. 5, the invention inputs the information obtained by sign identification into an Excel table, so that the identification result can be quantized. The method specifically comprises the following steps: and importing the road mileage information obtained in the acquisition into an Excel table, importing the identified traffic sign type into the Excel table, and importing the identified traffic sign position information into the Excel table.
The invention provides an automatic and efficient road traffic sign identification and classification method based on a Mask _ RCNN model. The method can automatically identify the traffic signs on the images, display the corresponding positions on the images and classify different traffic signs. The method can be applied to daily inspection of the expressway and recognition of surrounding objects by future unmanned driving. Meanwhile, the invention fully considers the reason of the error identification of the label, individually classifies the objects which are easy to interfere into one class in a targeted way, and eliminates the objects when the identification result is generated, thus greatly improving the accuracy of the identification result.
The identification and classification of the traffic sign have important significance for the auxiliary driving of the automobile or the future unmanned driving, and the invention provides an automatic and efficient method for the identification of the traffic sign and lays a foundation for the future unmanned driving research.
The above are preferred embodiments of the present invention, and all changes made according to the technical scheme of the present invention that produce functional effects do not exceed the scope of the technical scheme of the present invention belong to the protection scope of the present invention.
Claims (6)
1. A road traffic sign automatic identification and classification method is characterized by comprising the following steps:
step S1: collecting road images by adopting vehicle-mounted image collecting equipment;
step S2: screening out images with traffic signs from the acquired road images, marking the images, and constructing a data set required by Mask _ RCNN model training;
step S3: inputting the data set obtained in the step S2 into a Mask _ RCNN model for training to obtain a trained weight;
step S4: carrying out identification and classification on road signs on all the collected road images by using the weights trained in the step S3;
step S5: checking the generated result, carrying out secondary annotation on the image with poor recognition effect and retraining the weight;
step S6: and outputting the recognition result.
2. The method for automatically identifying and classifying road traffic signs according to claim 1, wherein the step S2 specifically comprises the following steps:
step S21: determining the traffic sign types: according to a mark information acquisition table of a highway maintenance center in Fujian province, and by combining acquired road images, the traffic mark types of roads are divided into six major categories of a notice mark, a prohibition mark, a warning mark, a road indication mark, an indication mark and a milepost;
and S22, labeling the road images through L abelme labeling software, selecting images with traffic signs from the acquired road images to label the images to obtain labeled images, labeling the traffic signs by using English names during labeling, wherein the English names corresponding to the notification signs, the prohibition signs, the warning signs, the road signs, the indication signs and the mileposts are respectively notice, promotion, warning, guide, indication and milestone, and if a plurality of signs of the same type appear in one road image, adding serial numbers after the names.
3. The method for automatically identifying and classifying road traffic signs according to claim 1, wherein in the step S3, the Mask _ RCNN model mainly comprises four main parts: (1) the main frame is a standard convolution neural network, the bottom layer detects low-level characteristics, and the high layer detects high-level characteristics; allowing the feature of each level to be combined with the high-level and low-level features during the propagation process of the convolutional neural network; (2) the region suggestion network RPN is a light neural network, scans images through a sliding window and finds a region where a target exists; (3) the ROI classifier and the boundary frame regressor are used for dividing the categories of the target objects on the basis of obtaining the target area based on the RPN; the ROI classifier can classify the region into a specific category and can also generate a background category; further fine-tuning the position and size of the bounding box by using a bounding box regressor to encapsulate the target; (4) dividing masks, wherein a mask branch is a convolution network, taking the positive regions selected by the ROI classifier as input, and generating masks of the positive regions; in the training process, the real mask is reduced to calculate a loss function, in the inference process, the predicted mask is enlarged to the size of an ROI frame to give a final mask result, and each target has one mask;
and (5) transmitting the image marked in the step (S2) into a Mask _ RCNN model for training to obtain a trained weight.
4. The method for automatically identifying and classifying road traffic signs according to claim 1, wherein in step S4, the weights obtained from the training in step S3 are saved in the file positions corresponding to the collected road images; setting a form of an output image, and if the content of the label is more than that of the traffic sign label set at the beginning, not displaying a mask of the label of the non-traffic sign on the image; and then, all the collected road images are imported into a model for training, and a result image with traffic sign category and position information is automatically generated.
5. The method for automatically identifying and classifying road traffic signs according to claim 1, wherein the step S5 specifically comprises the following steps:
step S51: checking the result generated in the step S4, and screening out the image with error identification;
step S52: re-labeling the data set for the false detection image appearing in step S51, specifically: returning to the step S2, adding two categories of windows and billboards on the basis of the original marking, and respectively naming the two categories as window and billboard; marking the two newly added classes, and retraining the marked image data; generating a new training result and a new recognition result;
step S53: and checking the new generated result until the recognition effect is in accordance with the expectation.
6. The method for automatically identifying and classifying road traffic signs according to claim 1, wherein in step S6, the step of inputting picture road traffic sign information into an Excel table comprises: and importing the road mileage information obtained in the acquisition into an Excel table, importing the identified traffic sign type into the Excel table, and importing the identified traffic sign position information into the Excel table.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010329229.5A CN111488854A (en) | 2020-04-23 | 2020-04-23 | Automatic identification and classification method for road traffic signs |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010329229.5A CN111488854A (en) | 2020-04-23 | 2020-04-23 | Automatic identification and classification method for road traffic signs |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111488854A true CN111488854A (en) | 2020-08-04 |
Family
ID=71811849
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010329229.5A Pending CN111488854A (en) | 2020-04-23 | 2020-04-23 | Automatic identification and classification method for road traffic signs |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111488854A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112381840A (en) * | 2020-11-27 | 2021-02-19 | 深源恒际科技有限公司 | Method and system for marking vehicle appearance parts in loss assessment video |
| CN112464737A (en) * | 2020-11-04 | 2021-03-09 | 浙江预策科技有限公司 | Road marking detection and identification method, electronic device and storage medium |
| CN112884705A (en) * | 2021-01-06 | 2021-06-01 | 西北工业大学 | Two-dimensional material sample position visualization method |
| CN113076800A (en) * | 2021-03-03 | 2021-07-06 | 惠州市博实结科技有限公司 | Road sign board detection method and device |
| CN113610043A (en) * | 2021-08-19 | 2021-11-05 | 海默潘多拉数据科技(深圳)有限公司 | Industrial drawing table structured recognition method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778740A (en) * | 2016-12-06 | 2017-05-31 | 北京航空航天大学 | A kind of TFDS non-faulting image detecting methods based on deep learning |
| CN108830277A (en) * | 2018-04-20 | 2018-11-16 | 平安科技（深圳）有限公司 | Training method, device, computer equipment and the storage medium of semantic segmentation model |
| US20190087673A1 (en) * | 2017-09-15 | 2019-03-21 | Baidu Online Network Technology (Beijing) Co., Ltd | Method and apparatus for identifying traffic light |
| CN110490151A (en) * | 2019-08-22 | 2019-11-22 | 昆明理工大学 | A kind of detection method based on Mask RCNN identification Terahertz safety check image suspicious object |
| CN110619279A (en) * | 2019-08-22 | 2019-12-27 | 天津大学 | Road traffic sign instance segmentation method based on tracking |
-
2020
- 2020-04-23 CN CN202010329229.5A patent/CN111488854A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778740A (en) * | 2016-12-06 | 2017-05-31 | 北京航空航天大学 | A kind of TFDS non-faulting image detecting methods based on deep learning |
| US20190087673A1 (en) * | 2017-09-15 | 2019-03-21 | Baidu Online Network Technology (Beijing) Co., Ltd | Method and apparatus for identifying traffic light |
| CN108830277A (en) * | 2018-04-20 | 2018-11-16 | 平安科技（深圳）有限公司 | Training method, device, computer equipment and the storage medium of semantic segmentation model |
| CN110490151A (en) * | 2019-08-22 | 2019-11-22 | 昆明理工大学 | A kind of detection method based on Mask RCNN identification Terahertz safety check image suspicious object |
| CN110619279A (en) * | 2019-08-22 | 2019-12-27 | 天津大学 | Road traffic sign instance segmentation method based on tracking |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112464737A (en) * | 2020-11-04 | 2021-03-09 | 浙江预策科技有限公司 | Road marking detection and identification method, electronic device and storage medium |
| CN112464737B (en) * | 2020-11-04 | 2022-02-22 | 浙江预策科技有限公司 | Road marking detection and identification method, electronic device and storage medium |
| CN112381840A (en) * | 2020-11-27 | 2021-02-19 | 深源恒际科技有限公司 | Method and system for marking vehicle appearance parts in loss assessment video |
| CN112884705A (en) * | 2021-01-06 | 2021-06-01 | 西北工业大学 | Two-dimensional material sample position visualization method |
| CN112884705B (en) * | 2021-01-06 | 2024-05-14 | 西北工业大学 | Two-dimensional material sample position visualization method |
| CN113076800A (en) * | 2021-03-03 | 2021-07-06 | 惠州市博实结科技有限公司 | Road sign board detection method and device |
| CN113610043A (en) * | 2021-08-19 | 2021-11-05 | 海默潘多拉数据科技(深圳)有限公司 | Industrial drawing table structured recognition method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111488854A (en) | Automatic identification and classification method for road traffic signs | |
| CN109977812B (en) | Vehicle-mounted video target detection method based on deep learning | |
| CN107729801B (en) | Vehicle color recognition system based on multitask deep convolution neural network | |
| CN110969160B (en) | License plate image correction and recognition method and system based on deep learning | |
| EP3806064A1 (en) | Method and apparatus for detecting parking space usage condition, electronic device, and storage medium | |
| CN102867417B (en) | Taxi anti-forgery system and taxi anti-forgery method | |
| CN103824081B (en) | Method for detecting rapid robustness traffic signs on outdoor bad illumination condition | |
| CN109816024A (en) | A kind of real-time automobile logo detection method based on multi-scale feature fusion and DCNN | |
| CN110837807A (en) | Identification method and system for fake-licensed vehicle | |
| CN107506763A (en) | A kind of multiple dimensioned car plate precise positioning method based on convolutional neural networks | |
| CN105844257A (en) | Early warning system based on machine vision driving-in-fog road denoter missing and early warning method | |
| CN104732227A (en) | Rapid license-plate positioning method based on definition and luminance evaluation | |
| CN111967313B (en) | Unmanned aerial vehicle image annotation method assisted by deep learning target detection algorithm | |
| CN112488083B (en) | Identification method, device and medium of traffic signal lamp based on key point extraction of hetmap | |
| CN104851288B (en) | Traffic light positioning method | |
| CN108875803A (en) | A kind of detection of harmful influence haulage vehicle and recognition methods based on video image | |
| CN113723377A (en) | Traffic sign detection method based on LD-SSD network | |
| CN110532961A (en) | A kind of semantic traffic lights detection method based on multiple dimensioned attention mechanism network model | |
| CN113269161A (en) | Traffic signboard detection method based on deep learning | |
| CN113888754A (en) | Vehicle multi-attribute identification method based on radar vision fusion | |
| CN115761513A (en) | Intelligent remote sensing identification method for mountain large landslide based on semi-supervised deep learning | |
| CN107392115B (en) | Traffic sign identification method based on hierarchical feature extraction | |
| CN110765900A (en) | DSSD-based automatic illegal building detection method and system | |
| CN115761223A (en) | Remote sensing image instance segmentation method by using data synthesis | |
| CN112818987B (en) | Method and system for identifying and correcting display content of electronic bus stop board |
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 |