CN112560670B - Deep learning-based traffic sign symbol and text detection and identification method and device - Google Patents
Deep learning-based traffic sign symbol and text detection and identification method and device Download PDFInfo
- Publication number
- CN112560670B CN112560670B CN202011476719.4A CN202011476719A CN112560670B CN 112560670 B CN112560670 B CN 112560670B CN 202011476719 A CN202011476719 A CN 202011476719A CN 112560670 B CN112560670 B CN 112560670B
- Authority
- CN
- China
- Prior art keywords
- traffic sign
- detection
- network
- recognition
- text
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/40—Document-oriented image-based pattern recognition
- G06V30/41—Analysis of document content
- G06V30/413—Classification of content, e.g. text, photographs or tables
-
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- 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
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Abstract
The invention relates to a method and a device for detecting and identifying traffic sign symbols and texts based on deep learning.A traffic sign identification network is established in a grid parameter searching mode, and a plurality of images are obtained through data enhancement transformation and are used as a plurality of inputs of a model, so that the accuracy of the model is further improved; the pre-training model parameters are used as initial parameters of the model, and a cascade method is adopted to detect and identify in stages, so that targeted data enhancement is performed under the condition of fewer samples, and the robustness of the model is improved; by using a channel pruning strategy, parameters are greatly reduced under the condition of slight loss of precision, so that the model speed is improved.
Description
Technical Field
The invention relates to the field of traffic sign symbol detection and identification and traffic sign text detection and identification, in particular to a method and a device for detecting and identifying traffic sign symbols and texts based on deep learning.
Background
The traffic sign on the road is a public sign with remarkable color and shape characteristics and is used for managing traffic, indicating driving direction and ensuring the facilities of road smoothness and driving safety. The traffic sign is an important carrier of traffic information and can accurately guide the traffic of vehicles and pedestrians, so that the traffic sign information can be accurately identified in time, which is important for traffic safety. At present, urban traffic in China is in an important node for upgrading and updating, and an intelligent traffic system becomes a main node for the development of each city. The method for efficiently identifying the road traffic signs, which can adapt to different scenes, is researched, and has important significance for an intelligent traffic system. Meanwhile, in recent years, the deep learning technology is rapidly developed, strong characteristic learning capability is shown, and a new idea is provided for the field.
Traffic signs can be classified into two types, a symbol type and a character type. The symbolic traffic sign only contains simple fixed numbers and symbolic marks, is simple in detection and identification difficulty and mainly comprises an indicating sign, a warning sign and a prohibition sign. The traffic sign in character form includes complex information such as numbers, characters and symbols, and the like, is difficult to detect and identify and mainly comprises a road sign, a tourist area sign and an auxiliary sign.
The traffic sign symbol detection algorithms are various and can be roughly divided into a traditional traffic sign symbol detection method and a deep learning-based traffic sign symbol detection method. The traditional traffic sign symbol detection method can be mainly divided into the following four methods: based on color feature detection, shape feature detection, template matching detection, color geometric fusion feature detection and the like, the traditional traffic sign detection method is not only to be improved in precision, but also has various limitations in application scenes and categories. With the development of deep learning, in the field of target detection, deep learning has dominated. A deep learning-based traffic sign detection method is a specific application of a mainstream target detection algorithm. Common target detection algorithms can be divided into region-based two-step methods and non-region-based one-step methods, the two-step method relies on the idea of RCNN more, and the one-step method relies on the idea of YOLOV1 more. R-CNN divides target detection into three parts: area recommendation; extracting characteristics; region classification and regression. YOLOV1 is the first single-step network, which directly uses the loss function to perform the detection and identification optimization after a series of convolutional layers, and does not need to select candidate frames, so the detection speed is greatly increased compared with the prior art. At present, the traffic sign recognition method can be divided into a traditional method and a deep learning-based method. The traditional recognition method mainly utilizes various image characteristics and a machine learning algorithm for classification. The traffic sign symbol recognition method based on deep learning uses the BP neural network at the earliest, and the mainstream method at that time is to extract the image characteristics and input the image characteristics to the BP neural network for traffic sign recognition. After the convolutional neural network appears, the correlation algorithm quickly takes the leading position in the traffic sign symbol identification field.
The detection difficulty of the traffic sign text is high, and only a few researches on the detection of the traffic sign text are carried out at present. The traditional traffic sign text detection is based on connected domain detection, but because Chinese characters have the characteristic of stroke separation, the method has inherent defects for Chinese character detection. Compared with the traditional text detection method, the traffic sign text detection based on deep learning can replace the traditional text saliency detection based on characters with the text region detection based on character strings, avoids a large number of processing steps such as bottom-up character clustering and segmentation, can well overcome the defects of sensitive illumination, shape deformation and the like, and has better robustness. Traffic sign text recognition and general text recognition have many similar properties, and therefore most studies of traffic sign text recognition use the method of open source OCR. With the development of deep learning in recent years, two major methods, which are respectively dominated by CTC and Attention, have been developed in the field of text recognition, for example, CRNN is a classical CTC-based method.
Generally, most of current-stage detection and identification methods for traffic sign symbols and texts do not combine with practical application requirements, do not consider similarity between traffic sign categories, and have the defects of low identification accuracy, long running time, incapability of meeting vehicle-mounted instantaneity and the like.
Disclosure of Invention
The invention aims to solve the technical problem of designing a method and a device for detecting and identifying traffic sign symbols and texts based on deep learning so as to solve the problems in the background technology.
The technical problem solved by the invention is realized by adopting the following technical scheme:
s1 creates a traffic sign symbol detection and identification data set and a traffic sign text detection and identification data set.
S2, adjusting the input size of YOLOv3, loading a pre-training model, training a traffic sign symbol detection model until the model loss converges, and reducing the model parameters by using a pruning strategy.
S3, the traffic sign symbol detection result is used as model input, and a traffic sign symbol recognition network is constructed: dense connection is used as a basic framework of network design, two hyperparameters of input size and layer number are set based on the idea of grid parameter search, and an optimal model structure is selected. On the basis of the optimal model structure, a single input is converted into a plurality of inputs through image enhancement transformation, the plurality of inputs are respectively connected with the optimal model structure in a back-to-back mode, and a traffic sign recognition model is trained until the model loss converges.
S4, the traffic sign symbol detection result is used as model input, and a traffic sign text detection network is constructed: aiming at the traffic sign text characteristics in the road scene, an EAST algorithm is improved, and a TSEAST traffic sign text detection algorithm is constructed.
S5, the traffic sign text detection result is used as a model input, and a traffic sign text recognition network is constructed: and fusing the ideas of the convolutional network and the cyclic neural network, and constructing a CRNN algorithm for recognizing the traffic sign text.
S6, the trained traffic sign symbol detection and recognition network and the traffic sign text detection and recognition network are cascaded to realize the detection and recognition of the traffic sign symbol and the text.
The invention has the advantages that: the traffic sign symbol has various types, and the traffic sign text is easily interfered by other character information. The invention fully considers the particularity of the detection and identification of the traffic sign symbol and the text, constructs the traffic sign identification network by a grid parameter searching mode, obtains a plurality of images as a plurality of inputs of the model by data enhancement transformation, and further improves the accuracy rate of the model; the pre-training model parameters are used as initial parameters of the model, and a cascade method is adopted to detect and identify in stages, so that targeted data enhancement is performed under the condition of fewer samples, and the robustness of the model is improved; by using a channel pruning strategy, parameters are greatly reduced under the condition of slight loss of precision, so that the model speed is improved.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a schematic diagram of a traffic sign symbol detection and recognition result.
FIG. 3 is a traffic sign text detection model combination module.
Fig. 4 is a schematic diagram of a traffic sign text detection result.
Fig. 5 is a schematic diagram of the detection and recognition results of the traffic sign symbol and text.
Detailed Description
In order to make the technical means, creation features, work flow, use method, achievement purpose and efficacy of the invention easy to understand, the invention is further explained below with reference to the accompanying drawings.
As shown in fig. 1, the embodiment of the invention discloses a method for detecting and identifying a traffic sign based on a cascaded deep neural network, which comprises the following steps:
s1, road video data is obtained, and a traffic sign detection and identification data set is made.
Respectively making a traffic sign detection and identification data set, and respectively adding 6: 2: 2, dividing the training set, the verification set and the test set into three data sets.
The traffic sign detection data set is produced using a VOC format. The road video data is different from the image data, and the image similarity in the same video is high, so that the pictures extracted from the same road video are divided into the same data set, and the samples which are similar to the training set in polarity do not exist in the verification set and the test set as far as possible, thereby improving the verification capability of the generalization of the model.
After the traffic sign detection data set is manufactured, the traffic signs in the image are cut according to the xml information, and the traffic signs are placed into different folders according to categories to be used as traffic sign identification data sets.
S2, constructing a traffic sign symbol detection model:
the traffic sign in the road video data set usually occupies a small area and belongs to the problem of input small target detection. Under the same condition, the detection effect of the model on the small target can be greatly improved by enlarging the size of the input image, but the detection time is increased at the same time. The two are weighed according to practical requirements, and the invention adopts 960 multiplied by 960 resolution as input size.
In order to enhance the experimental effect and generalization capability of the traffic sign detection model, weights trained on the COCO are used as initialization parameters of the YOLOV3 model. Firstly freezing all layers of YOLOV3 except the last three layers, using an Adam optimizer, training 20 iteration cycles at the learning rate of 0.01, then releasing the freezing, using the Adam optimizer, training 50 iteration cycles at the learning rate of 0.001, adopting an early stopping learning strategy, and stopping the iteration when the evaluation index is not changed for 10 continuous iteration cycles.
The trained YOLOV3 model parameters are large, so network slimming reduction model parameters are used. The core idea of network slimming is that L1 is applied to a scaling factor of a batch normalization layer in a standardized manner, so that the scaling factor tends to zero, finally, a threshold value is set for the scaling factor to perform channel pruning to delete unimportant channels, the precision after pruning may be temporarily reduced, but the precision can be adjusted back by fine adjustment of the network after pruning, the process is repeatable, and the precision error is guaranteed not to exceed five per thousand by setting parameters and repetition times of sparse training. Finally, the detection of three traffic sign symbols of warning, prohibition and indication and the traffic sign in character type is realized with high precision and high speed.
S3, constructing a traffic sign symbol recognition model:
the basic module of the traffic sign symbol recognition model uses the concept of dense connection, and uses the concept of grid parameter search to search three hyper-parameters, namely the number of modules, the number of convolution layers of each module and the input size, so as to finally construct a 40-layer feature extraction module, and the input size is selected to be 32 multiplied by 32.
On the basis of the optimal model, three additional data sets are obtained through three data enhancement transformations of contrast enhancement, histogram equalization and local histogram equalization, and the three data sets are added with the original image data set and are respectively used for training the traffic sign symbol recognition model to obtain four models. During prediction, the predicted probability results of all categories are averaged to obtain the final average prediction result as an output result.
S4, constructing a traffic sign text detection model:
the EAST algorithm comprises two phases: the method comprises a full convolution network stage and a non-maximum suppression fusion stage, wherein the full convolution network stage is used for directly generating a text candidate region, and the non-maximum suppression fusion stage is used for screening a final text region. The full convolution network stage can be further specifically divided into a feature extraction branch, a feature fusion branch and an output layer. The traffic sign text under the natural scene has the characteristics of inclination change, perspective change, bending change, shielding, large length-width ratio and the like. The combination module uses irregular convolution, such as 1 × 5 convolution and 5 × 1 convolution, so as to generate a rectangular receptive field, better match a traffic sign text with a large length-width ratio, and simultaneously uses deformable convolution, so that the model has good geometric transformation capability to deal with challenges in aspects of target scale, posture, partial deformation and the like, so as to extract better semantic information, and the specific structure is shown in fig. 3.
S5, constructing a traffic sign text recognition model:
the convolutional recurrent neural network mainly comprises the following three parts: 1) the convolutional neural network layer is mainly responsible for extracting the features of the input picture so as to obtain a feature sequence; 2) the cyclic neural network layer is mainly responsible for predicting the characteristic sequences one by one; 3) and the transcription layer is mainly responsible for converting and sorting the sequence prediction result to obtain a final prediction result. Because the traffic sign text recognition data set under the natural scene is seriously deficient, in the training process, firstly, model training is carried out on the artificially synthesized traffic sign text recognition training data set, and a model with the highest precision is selected on a corresponding verification set; and then carrying out fine adjustment on a traffic sign text recognition training data set in a natural scene, loading weights, continuing training at a lower learning rate, adopting an early-stopping learning strategy, stopping iteration when evaluation indexes are not changed for 10 continuous iteration cycles, and selecting a model with the highest precision from a traffic sign text recognition verification set in the natural scene.
S6 cascade training traffic sign symbol detection and recognition network and traffic sign text detection and recognition network
The invention relates to a conventional deep learning target detection and identification algorithm which simultaneously completes the detection and identification of targets in the same network, and because of the defects of large quantity of traffic sign types and large similarity among the types, the invention takes the result obtained by a traffic sign symbol detection network as the input of a traffic sign symbol identification network and a traffic sign text detection network, and then takes the result of the traffic sign text detection as the input of the traffic sign text identification network, and realizes the detection and identification of traffic sign symbols and texts by adopting a network cascade mode.
Through tests, the method has a good application effect on the detection and identification problems of small targets such as traffic signs, can realize high detection and identification accuracy of traffic sign symbols and texts, and has a comprehensive identification effect shown in figure 5.
Based on the same idea, the invention also designs a traffic sign symbol and text detection and recognition device based on deep learning, which is characterized in that: comprises the steps of (a) preparing a mixture of a plurality of raw materials,
the data acquisition module is used for acquiring and manufacturing a traffic sign symbol detection and identification data set and a traffic sign text detection and identification data set;
the traffic sign symbol detection module trains a traffic sign symbol detection model by using YOLOV 3;
and the traffic sign symbol recognition module is used for inputting the traffic sign symbol detection result as a model to construct a traffic sign symbol recognition network: setting two super parameters of input size and layer number based on the idea of grid parameter search by taking dense connection as a basic framework of network design, and selecting an optimal model structure; on the basis of the optimal model structure, converting a single input into a plurality of inputs through image enhancement transformation, and respectively connecting the plurality of inputs with the optimal model structure in sequence to train a traffic sign symbol recognition model;
the traffic sign symbol recognition module is used for inputting a traffic sign symbol detection result as a model, constructing a traffic sign text detection network, constructing a TSEAST traffic sign text detection model by using an improved EAST algorithm and training;
the traffic sign text detection module is used for inputting a traffic sign text detection result as a model, and constructing a traffic sign text recognition network by adopting a CRNN (CrNN network) for detecting the traffic sign text;
and the traffic sign text recognition module is used for cascading the trained traffic sign symbol detection and recognition network and the trained traffic sign text detection and recognition network to realize the detection and recognition of the traffic sign symbols and texts.
Claims (8)
1. A method for detecting and identifying traffic sign symbols and texts based on deep learning is characterized in that: the method comprises the following steps:
s1, making a traffic sign symbol detection and identification data set and a traffic sign text detection and identification data set;
s2 training the traffic sign symbol detection network using YOLOV 3;
s3, the traffic sign symbol detection result is used as network input to construct a traffic sign symbol recognition network: setting two super parameters of input size and layer number based on the idea of grid parameter search by taking dense connection as a basic framework of network design, and selecting an optimal model structure; on the basis of the optimal model structure, converting a single input into a plurality of inputs through image enhancement transformation, and respectively connecting the plurality of inputs with the optimal model structure to train a traffic sign symbol recognition network;
s4, inputting the traffic sign symbol detection result as a network, constructing a traffic sign text detection network, specifically constructing the traffic sign text detection network by adopting an improved EAST algorithm, namely a TSEAST algorithm, and training, wherein the improved EAST algorithm is a feature extraction module based on irregular convolution and deformable convolution before a model output layer;
s5, inputting the traffic sign text detection result as a network, and constructing a traffic sign text recognition network, specifically adopting a CRNN traffic sign text recognition network;
s6, the trained traffic sign symbol detection and recognition network and the traffic sign text detection and recognition network are cascaded to realize the detection and recognition of the traffic sign symbol and the text.
2. The detection and identification method of claim 1, wherein: the data adopted in step S1 is road video data, image data is extracted frame by frame, a traffic sign detection data set and a traffic sign identification data set are created, and the traffic sign detection data set adopts a VOC data format.
3. The detection and identification method of claim 1, wherein: the training model input image resolution size in step S2 is 960 × 960.
4. The detection and identification method according to claim 1, wherein: in step S2, channel pruning is used to perform model fine tuning to reduce model parameters.
5. The detection and identification method of claim 1, wherein: the step S2 of training the traffic sign symbol detection network by using YOLOv3 specifically includes:
using the weights trained on the COCO as initialization parameters of the network; firstly freezing all layers of YOLOV3 except the last three layers, using an Adam optimizer, training 20 iteration cycles at the learning rate of 0.01, then releasing the freezing, using the Adam optimizer, training 50 iteration cycles at the learning rate of 0.001, adopting an EarlyStopping learning strategy, and stopping the iteration when the evaluation index is not changed for 5 continuous iteration cycles.
6. The detection and identification method of claim 1, wherein: in step S3, the input size is set to 32 × 32, and the number of layers is 40.
7. The detection and identification method of claim 1, wherein: in the process of training the CRNN network in step S5, firstly, network training is performed on the artificially synthesized traffic sign text recognition training data set, a network with the highest accuracy is selected on the corresponding verification set, then, training fine tuning is performed on the traffic sign text recognition training set in a natural scene, and a network with the highest accuracy is selected on the traffic sign text recognition verification data set in the natural scene.
8. A detection and recognition device of traffic sign symbols and texts based on deep learning is characterized in that: comprises that
The data acquisition module is used for acquiring and manufacturing a traffic sign symbol detection and identification data set and a traffic sign text detection and identification data set;
the traffic sign symbol detection module trains a traffic sign symbol detection network by using YOLOV 3;
and the traffic sign symbol recognition module takes the detection result of the traffic sign symbol as network input to construct a traffic sign symbol recognition network: setting two super parameters of input size and layer number based on the idea of grid parameter search by taking dense connection as a basic framework of network design, and selecting an optimal model structure; on the basis of the optimal model structure, converting a single input into a plurality of inputs through image enhancement transformation, and respectively connecting the plurality of inputs with the optimal model structure to train a traffic sign symbol recognition network;
the traffic sign symbol recognition module is used for inputting a traffic sign symbol detection result as a network, constructing a traffic sign text detection network, specifically adopting an improved EAST algorithm, namely a TSEAST algorithm, constructing the traffic sign text detection network, and training, wherein the improved EAST algorithm is a feature extraction module based on irregular convolution and deformable convolution before a network output layer;
the traffic sign text detection module is used for inputting a traffic sign text detection result as a network, adopting a CRNN (CrNN) network to construct a traffic sign text recognition network and detecting the traffic sign text;
and the traffic sign text recognition module is used for cascading the trained traffic sign symbol detection and recognition network and the traffic sign text detection and recognition network to realize the detection and recognition of the traffic sign symbols and the texts.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011476719.4A CN112560670B (en) | 2020-12-15 | 2020-12-15 | Deep learning-based traffic sign symbol and text detection and identification method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011476719.4A CN112560670B (en) | 2020-12-15 | 2020-12-15 | Deep learning-based traffic sign symbol and text detection and identification method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112560670A CN112560670A (en) | 2021-03-26 |
| CN112560670B true CN112560670B (en) | 2022-08-16 |
Family
ID=75063680
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011476719.4A Active CN112560670B (en) | 2020-12-15 | 2020-12-15 | Deep learning-based traffic sign symbol and text detection and identification method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112560670B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113284164A (en) * | 2021-05-19 | 2021-08-20 | 中国农业大学 | Shrimp swarm automatic counting method and device, electronic equipment and storage medium |
| CN113239217B (en) * | 2021-06-04 | 2024-02-06 | 图灵深视(南京)科技有限公司 | Image index library construction method and system, and image retrieval method and system |
| CN113435437A (en) * | 2021-06-24 | 2021-09-24 | 随锐科技集团股份有限公司 | Method and device for identifying state of switch on/off indicator and storage medium |
| CN113963329B (en) * | 2021-10-11 | 2022-07-05 | 浙江大学 | Digital traffic sign detection and identification method based on double-stage convolutional neural network |
| CN113947764B (en) * | 2021-12-06 | 2022-03-08 | 腾讯科技（深圳）有限公司 | Image processing method, device, equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103390167A (en) * | 2013-07-18 | 2013-11-13 | 奇瑞汽车股份有限公司 | Multi-characteristic layered traffic sign identification method |
| CN110348396A (en) * | 2019-07-15 | 2019-10-18 | 南京信息工程大学 | A kind of road top text traffic sign and device based on deep learning |
| CN111931683A (en) * | 2020-08-25 | 2020-11-13 | 腾讯科技（深圳）有限公司 | Image recognition method, image recognition device and computer-readable storage medium |
| CN112016514A (en) * | 2020-09-09 | 2020-12-01 | 平安科技（深圳）有限公司 | Traffic sign identification method, device, equipment and storage medium |
-
2020
- 2020-12-15 CN CN202011476719.4A patent/CN112560670B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103390167A (en) * | 2013-07-18 | 2013-11-13 | 奇瑞汽车股份有限公司 | Multi-characteristic layered traffic sign identification method |
| CN110348396A (en) * | 2019-07-15 | 2019-10-18 | 南京信息工程大学 | A kind of road top text traffic sign and device based on deep learning |
| CN111931683A (en) * | 2020-08-25 | 2020-11-13 | 腾讯科技（深圳）有限公司 | Image recognition method, image recognition device and computer-readable storage medium |
| CN112016514A (en) * | 2020-09-09 | 2020-12-01 | 平安科技（深圳）有限公司 | Traffic sign identification method, device, equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| Cascaded Segmentation-Detection Networks for Text-Based Traffic Sign Detection;Yingying Zhu 等;《Journals&Magazines》;20171231;第19卷(第01期);全文 * |
| 交通标志检测与识别研究;朱盈盈;《中国博士学位论文全文数据库》;20190515(第05期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112560670A (en) | 2021-03-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112560670B (en) | Deep learning-based traffic sign symbol and text detection and identification method and device | |
| CN103761531B (en) | The sparse coding license plate character recognition method of Shape-based interpolation contour feature | |
| CN106022300B (en) | Traffic sign recognition method and system based on cascade deep study | |
| CN109614979B (en) | Data augmentation method and image classification method based on selection and generation | |
| CN105160309B (en) | Three lanes detection method based on morphological image segmentation and region growing | |
| CN112069940B (en) | Cross-domain pedestrian re-identification method based on staged feature learning | |
| CN110175613A (en) | Street view image semantic segmentation method based on Analysis On Multi-scale Features and codec models | |
| CN107122776A (en) | A kind of road traffic sign detection and recognition methods based on convolutional neural networks | |
| CN104778470B (en) | Text detection based on component tree and Hough forest and recognition methods | |
| CN103870803A (en) | Vehicle license plate recognition method and system based on coarse positioning and fine positioning fusion | |
| CN111368754B (en) | Airport runway foreign matter detection method based on global context information | |
| CN106257496B (en) | Mass network text and non-textual image classification method | |
| CN111767927A (en) | Lightweight license plate recognition method and system based on full convolution network | |
| CN105718866A (en) | Visual target detection and identification method | |
| CN110443805A (en) | A kind of semantic segmentation method spent closely based on pixel | |
| CN112990282B (en) | Classification method and device for fine-granularity small sample images | |
| CN110866388A (en) | Publishing PDF layout analysis and identification method based on mixing of multiple neural networks | |
| CN111191608A (en) | Improved traffic sign detection and identification method based on YOLOv3 | |
| CN111738300A (en) | Optimization algorithm for detecting and identifying traffic signs and signal lamps | |
| CN108491828B (en) | Parking space detection system and method based on level pairwise similarity PVAnet | |
| CN111126401A (en) | License plate character recognition method based on context information | |
| Sun et al. | Deep learning based pedestrian detection | |
| CN112131996B (en) | Road side image multi-scale pedestrian rapid detection method based on channel separation convolution | |
| CN115424243A (en) | Parking stall number identification method, equipment and medium based on yolov5-shufflenetv2 | |
| CN113378838A (en) | Method for detecting text region of nameplate of mutual inductor based on deep learning |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |