CN111126331A - Real-time guideboard detection method combining object detection and object tracking - Google Patents
Real-time guideboard detection method combining object detection and object tracking Download PDFInfo
- Publication number
- CN111126331A CN111126331A CN201911405434.9A CN201911405434A CN111126331A CN 111126331 A CN111126331 A CN 111126331A CN 201911405434 A CN201911405434 A CN 201911405434A CN 111126331 A CN111126331 A CN 111126331A
- Authority
- CN
- China
- Prior art keywords
- guideboard
- detection
- network
- coordinates
- frame
- 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20104—Interactive definition of region of interest [ROI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30248—Vehicle exterior or interior
- G06T2207/30252—Vehicle exterior; Vicinity of vehicle
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a real-time guideboard detection method combining object detection and object tracking, which is characterized in that a guideboard detection network is constructed for positioning the coordinate position of a guideboard in a picture, and an initial frame is extracted from a to-be-detected video containing the guideboard and input into a trained guideboard detection network; and then obtaining the guideboard detection coordinates of the initial frame by a non-maximum suppression (NMS) method and a confidence threshold method. Inputting the guideboard detection coordinates into a KCF algorithm to obtain the guideboard tracking coordinates of the next frame in the video to be detected, continuously tracking the video to be detected by a fixed frame number according to the requirement, taking the last frame continuously tracked as a new initial frame, and repeating the steps to obtain whether each frame in the video to be detected contains the guideboard and the coordinates of the guideboard. The method combines the accuracy of the detection method and simultaneously utilizes the rapidity of the tracking algorithm.
Description
Technical Field
The invention belongs to the field of machine vision application, and particularly relates to a real-time guideboard detection method combining object detection and object tracking.
Background
The research of the guideboard detection system starts from the beginning of the last century, and the guideboard detection system also becomes a popular research direction with the attention of intelligent driving and automatic driving. The guideboard detection method based on machine learning has the advantages of low cost, high accuracy, strong real-time performance and easiness in implementation, and is a necessary topic in the field of unmanned driving. Due to the complexity of traffic environment, the variability of illumination weather and the difference of traffic signs in various countries and regions, the current guideboard detection is still in a preliminary field application stage, the types of detected signs are limited, and the detection precision and efficiency have a great space for improvement, so high-precision and low-delay guideboard detection is still to be developed.
The current guideboard detection mainly has two main modes: one is a double-step detection mode, and the other is a single-step detection mode. The double-step detection means that the candidate area containing the guideboard is roughly detected, and then the accurate coordinate of the guideboard is accurately positioned, and the algorithm mainly comprises a FasterRCNN algorithm and the like. The single-step detection refers to a method of inputting a picture into a neural network to directly obtain coordinates, such as an SSD algorithm and a YOLO algorithm. The former network has complex and large structure and slower calculation speed, but has the advantages of high precision, and the latter network is small and fast, but the precision is far lower than that of the former network, and the design requirement on a neural network is higher.
Disclosure of Invention
The invention aims to provide a real-time guideboard detection method combining the advantages of object detection and object tracking aiming at the technical defects in the existing guideboard detection.
The purpose of the invention is realized by the following technical scheme: a real-time guideboard detection method combining object detection and object tracking comprises the following steps:
(1) collecting a video of an automobile in the advancing process; extracting video frames as a training picture set according to a fixed frame number interval, marking the coordinates of the guideboard in the picture by using a rectangular frame for the training picture containing the guideboard, and storing the coordinates in an xml file; all the xml files and the training pictures are used as a training set;
(2) and (3) constructing a guideboard detection network for positioning the coordinate position of the guideboard in the picture, inputting the training picture set in the step (1), and outputting the coordinates and the confidence coefficient of the guideboard predicted in each picture.
The basic network of the guideboard detection network is a network with a fully-connected layer removed by an inclusion V4 model pre-trained on an ImageNet data set, a region candidate network (RPN) is used behind the basic network to obtain a feature map containing guideboard candidate region coordinates, the feature map is subjected to region-of-interest pooling (ROIPooling) operation to obtain feature maps of candidate regions with uniform sizes, and each feature map is input into a positioning and classifying network to obtain accurate coordinates and confidence degrees of the candidate regions in an original image.
(3) Inputting the training set obtained in the step (1) into the constructed guideboard detection network for training, and storing guideboard detection network parameters to obtain the trained guideboard detection network;
(4) and (3) extracting an initial frame from the video to be detected containing the guideboard, inputting the initial frame into the guideboard detection network trained in the step (3), and obtaining the guideboard detection coordinate of the initial frame by a non-maximum suppression (NMS) and confidence threshold method.
(5) Inputting the guideboard detection coordinates output in the step (4) into a KCF algorithm to obtain guideboard tracking coordinates of the next frame in the video to be detected, and continuously tracking the video to be detected by a fixed frame number according to the requirement;
(6) and (4) taking the last frame continuously tracked in the step (5) as a new initial frame, and repeating the steps (4) and (5) to obtain whether each frame in the video to be detected contains the guideboard and the coordinates of the guideboard.
Further, in the step (1), labelImg software is adopted for coordinate labeling of the guideboard in the training picture.
Further, in the step (2), the positioning and classifying network is a 5-layer network, and the two networks share a four-layer network, which is a full connection layer with 1024 nodes, a relu activation layer, a full connection layer with 1024 nodes, and a relu activation layer; the fifth layer network of the classification network is a full connection layer with the node number of 2, and then output results are converted into a probability form through softmax operation; the fifth layer of the positioning network is a fully connected layer with the number of nodes being 8.
Further, in the step (2), the constructed guideboard detection network is pre-trained on the COCO data set, and the pre-trained network parameter result is taken as the initial parameters of the guideboard detection network.
Further, in the step (3), the training process of the guideboard detection network specifically includes: and (3) calculating a position loss function according to the guideboard detection coordinates obtained in the step (2) and the guideboard coordinates marked in the step (1), and calculating a classification loss function according to the guideboard confidence obtained in the step (2). And adding the two loss functions, deriving the total loss function, and updating the numerical value of each node in the network by a random gradient descent method with Momentum correction (Momentum) added to obtain the guideboard detection network parameter.
Further, in the step (3), random image augmentation is performed on the trained image, and the augmentation operation includes: random horizontal flipping, random picture scaling, random cutting, random rotation, and the like.
Further, in the step (5), in the tracking stage, the tracked coordinates are judged every fixed frame, the coordinates of the guideboard are tracked only when the same guideboard is detected in more than a fixed number of continuous frames, and the number of the tracked continuous frames is set according to requirements.
Further, in the step (6), the last frame of continuous tracking is used as a new initial frame, the new guideboard detection coordinate is obtained in the step (4), the new guideboard detection coordinate and the guideboard tracking coordinate of the last frame are integrated through the NMS algorithm to obtain the guideboard coordinate in the current frame, and the guideboard coordinate is input to the KCF algorithm in the step (5).
The invention has the beneficial effects that: the invention effectively combines the object detection algorithm and the object tracking algorithm, and builds the real-time guideboard detection network. In addition, in the process of training the guideboard detection model, the adjustment of the network weight is guided by a random gradient descent method adopting momentum correction. The method deletes effective guideboard tracking coordinates in a tracking module, and obtains finally predicted guideboard coordinates by adopting an NMS algorithm when integrating guideboard detection coordinates and guideboard tracking coordinates.
Drawings
FIG. 1 is a flow chart of guideboard detection convolutional neural network training;
FIG. 2 is a flow chart of a real-time detection algorithm;
fig. 3 is a schematic diagram of a guideboard label.
Detailed Description
The invention is described in further detail below with reference to the figures and specific examples.
The invention realizes high-precision real-time guideboard detection by building and training a guideboard detection convolutional neural network model and combining the guideboard detection convolutional neural network model with a guideboard tracking algorithm, and is characterized by building and training an efficient detection network architecture and integrating the coordinates of the detection and tracking guideboards. First, real training data is generated for training of the detection neural network. Secondly, a reasonable classification and positioning neural network is needed to be designed for road sign prediction, and the adjustment of the network weight is guided by a loss function and a momentum-corrected random gradient descent method. And demonstrates the overall accuracy and real-time of the model with examples. As shown in fig. 1, the main steps of the technical scheme adopted by the invention are as follows:
(1) and acquiring a video of the automobile in the advancing process through the camera. And extracting video frames with a constant frame number to serve as a training picture set, marking the coordinates of the guideboard in the training picture by using a rectangular frame through labelImg software, and storing the coordinates in an xml file. And taking the xml file and the training picture as a training set.
(2) And (3) building a guideboard detection depth network to position the coordinate position of the guideboard in the picture, inputting the training image set in the step (1), and outputting the coordinates and the confidence coefficient of the guideboard in each picture.
Guideboard detection network architecture: the basic network of the detection network is a network except a full connection layer of an inclusion V4 model pre-trained on an ImageNet data set, then a region candidate network (RPN) is utilized to obtain candidate region coordinates which are guideboards in an input picture, a feature map of the candidate region is subjected to region of interest pooling (ROIPooling) operation to obtain feature maps of candidate regions with uniform sizes, and all the feature maps are input into a positioning and classifying network to obtain accurate coordinates and confidence degrees of the candidate regions in an original image. The positioning and classifying network is a 5-layer network, the positioning and classifying network and the classifying network share four-layer networks, namely a full connection layer with 1024 nodes, a relu activation layer, a full connection layer with 1024 nodes and a relu activation layer. The fifth layer of the classified network is a fully connected layer with the node number of 2, and then output results are converted into a probability form through softmax operation. The fifth layer of the positioning network is a fully connected layer with the number of nodes being 8.
(3) The method is a training stage of the guideboard detection network, firstly, the constructed guideboard detection network is pre-trained on a COCO data set, and a pre-trained network parameter result is taken as an initial parameter of the guideboard detection network. And carrying out random image augmentation on the trained image, wherein the augmentation operation comprises the following steps: random horizontal flipping, random picture scaling, random cutting, random rotation, and the like. And then inputting the road sign into a road sign detection network, and training to obtain road sign detection network parameters. And (3) calculating a position loss function according to the guideboard detection coordinates obtained in the step (2) and the guideboard coordinates marked in the step (1), and calculating a classification loss function according to the guideboard confidence obtained in the step (2). And adding the two loss functions, after the total loss function is derived, updating the numerical value of each node in the network by a random gradient descent method of adding Momentum correction (Momentum), stopping updating the network weight when the loss function tends to be stable or reaches a set step number, and storing the current network weight. This experiment sets up training step number and is 200000 steps, and when 200000 steps, the change of loss function is no longer than 1%, tends to stable.
(4) And (3) extracting an initial frame from the video to be detected containing the guideboard, inputting the initial frame into the guideboard detection network trained in the step (3), and obtaining the guideboard detection coordinate of the initial frame by a non-maximum suppression (NMS) and confidence threshold method. The confidence threshold method is that the coordinates of the guideboard are adopted only when the confidence of the detected guideboard is greater than a certain threshold, otherwise, the confidence is discarded. Experiments show that when the threshold value is set to be 0.5, the network detection effect is optimal.
(5) Inputting the detection guideboard detection coordinates output in the step (4) into a KCF algorithm to obtain guideboard tracking coordinates of the next frame in the video to be detected, and continuously tracking the video to be detected by a fixed frame number according to the requirement; the accuracy and the speed of the method are optimal when the continuous tracking fixed frame number is 5 frames; in the tracking stage, the tracked coordinates are judged every fixed frame, and the guideboard coordinates are tracked only when the same guideboard is detected in more than a fixed number of continuous frames. It has been found through experimentation that the tracking algorithm works best when a fixed number of consecutive frames is set to 2.
(6) And (4) taking the last frame continuously tracked in the step (5) as a new initial frame, obtaining a new guideboard detection coordinate in the step (4), integrating the new guideboard detection coordinate and the guideboard tracking coordinate of the last frame through an NMS (network management system) algorithm to obtain the guideboard coordinate in the current frame, inputting the guideboard coordinate into the KCF algorithm in the step (5), and obtaining whether each frame in the video to be detected contains a guideboard and the coordinate of the guideboard.
Example (b):
mean Average Precision (mAP) of the PASCAL VOC game and the number of Frames Per Second (FPS) that the algorithm can handle are taken as the criteria for evaluating the algorithm.
Due to the lack of a public guideboard detection data set, the data set adopted by the data of the embodiment is taken from the following four aspects; dataset one, hundred degrees apollospace dataset. Data set two, the TT100K data set in qinghua tengchun. And thirdly, downloading pictures containing the guideboards from the hundred-degree pictures. And a data set IV, wherein the specific pictures and the number of the guideboards are extracted from the shot road condition video as shown in the table below.
| apollospace data set | TT100K data set | Baidu picture | Real shooting road condition video picture | Total value | |
| Number of pictures containing guideboard | 3370 | 1945 | 968 | 120 | 6403 |
| Number of guideboards | 4452 | 4629 | 2236 | 175 | 11492 |
| Number of pictures | 4022 | 4449 | 968 | 408 | 9847 |
The training stage adopts the data of the first three data sets, and the testing stage adopts the last real-shot road condition video as a testing object. The experimental result of the model reached an mAP of 0.75, which was 0.48 of the mAP of the comparative single-step detection algorithm SSD. The detection rate of the model reaches 32FPS, and the real-time requirement is met.
The above-described embodiments are intended to illustrate rather than to limit the invention, and any modifications and variations of the present invention are within the spirit of the invention and the scope of the appended claims.
Claims (8)
1. A real-time guideboard detection method combining object detection and object tracking is characterized by comprising the following steps:
(1) collecting a video of an automobile in the advancing process; extracting video frames as a training picture set according to a fixed frame number interval, marking the coordinates of the guideboard in the picture by using a rectangular frame for the training picture containing the guideboard, and storing the coordinates in an xml file; all the xml files and the training pictures are used as a training set;
(2) and (3) constructing a guideboard detection network for positioning the coordinate position of the guideboard in the picture, inputting the training picture set in the step (1), and outputting the coordinates and the confidence coefficient of the guideboard predicted in each picture.
The basic network of the guideboard detection network is a network with a fully-connected layer removed by an inclusion V4 model pre-trained on an ImageNet data set, a region candidate network (RPN) is used behind the basic network to obtain a feature map containing guideboard candidate region coordinates, the feature map is subjected to region-of-interest pooling (ROIPooling) operation to obtain feature maps of candidate regions with uniform sizes, and each feature map is input into a positioning and classifying network to obtain accurate coordinates and confidence degrees of the candidate regions in an original image.
(3) Inputting the training set obtained in the step (1) into the constructed guideboard detection network for training, and storing guideboard detection network parameters to obtain the trained guideboard detection network;
(4) and (3) extracting an initial frame from the video to be detected containing the guideboard, inputting the initial frame into the guideboard detection network trained in the step (3), and obtaining the guideboard detection coordinate of the initial frame by a non-maximum suppression (NMS) and confidence threshold method.
(5) Inputting the guideboard detection coordinates output in the step (4) into a KCF algorithm to obtain guideboard tracking coordinates of the next frame in the video to be detected, and continuously tracking the video to be detected by a fixed frame number according to the requirement;
(6) and (4) taking the last frame continuously tracked in the step (5) as a new initial frame, and repeating the steps (4) and (5) to obtain whether each frame in the video to be detected contains the guideboard and the coordinates of the guideboard.
2. The method for detecting the guideboard in real time by combining the object detection and the object tracking as claimed in claim 1, wherein in the step (1), labelImg software is used for marking the coordinates of the guideboard in the training picture.
3. The method for detecting the guideboard in real time by combining the object detection and the object tracking according to claim 1, wherein in the step (2), the positioning and classifying network is a 5-layer network, and the positioning and classifying network and the classifying network share four layers, namely a full-connection layer with 1024 nodes, a relu active layer, a full-connection layer with 1024 nodes and a relu active layer; the fifth layer network of the classification network is a full connection layer with the node number of 2, and then output results are converted into a probability form through softmax operation; the fifth layer of the positioning network is a fully connected layer with the number of nodes being 8.
4. The method according to claim 1, wherein in the step (2), the constructed road sign detection network is pre-trained on the COCO data set, and the pre-trained network parameter result is taken as the initial parameter of the road sign detection network.
5. The real-time guideboard detection method combining object detection and object tracking according to claim 1, wherein in step (3), the training process of the guideboard detection network specifically comprises: and (3) calculating a position loss function according to the guideboard detection coordinates obtained in the step (2) and the guideboard coordinates marked in the step (1), and calculating a classification loss function according to the guideboard confidence obtained in the step (2). And adding the two loss functions, deriving the total loss function, and updating the numerical value of each node in the network by a random gradient descent method with Momentum correction (Momentum) added to obtain the guideboard detection network parameter.
6. The method for real-time guideboard detection combined with object detection and object tracking according to claim 1, wherein in step (3), the training image is subjected to random image augmentation, and the augmentation operation includes: random horizontal flipping, random picture scaling, random cutting, random rotation, and the like.
7. The method as claimed in claim 1, wherein in the step (5), the tracked coordinates are determined at regular intervals during the tracking stage, and the coordinates of the guideboard are tracked only when the same guideboard is detected in more than a fixed number of consecutive frames, and the number of the tracked consecutive frames is set according to the requirement.
8. The real-time guideboard detection method combining object detection and object tracking according to claim 1, characterized in that in step (6), the last frame of continuous tracking is used as a new initial frame, new guideboard detection coordinates are obtained in step (4), the new guideboard detection coordinates and the guideboard tracking coordinates of the last frame are integrated by NMS algorithm to obtain the guideboard coordinates in the current frame, and the guideboard coordinates are input to KCF algorithm in step (5).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911405434.9A CN111126331A (en) | 2019-12-30 | 2019-12-30 | Real-time guideboard detection method combining object detection and object tracking |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911405434.9A CN111126331A (en) | 2019-12-30 | 2019-12-30 | Real-time guideboard detection method combining object detection and object tracking |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111126331A true CN111126331A (en) | 2020-05-08 |
Family
ID=70506029
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911405434.9A Pending CN111126331A (en) | 2019-12-30 | 2019-12-30 | Real-time guideboard detection method combining object detection and object tracking |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111126331A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111930877A (en) * | 2020-09-18 | 2020-11-13 | 蘑菇车联信息科技有限公司 | Map guideboard generation method and electronic equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170091950A1 (en) * | 2015-09-30 | 2017-03-30 | Fotonation Limited | Method and system for tracking an object |
| CN107316001A (en) * | 2017-05-31 | 2017-11-03 | 天津大学 | Small and intensive method for traffic sign detection in a kind of automatic Pilot scene |
| CN108320297A (en) * | 2018-03-09 | 2018-07-24 | 湖北工业大学 | A kind of video object method for real time tracking and system |
| CN110490073A (en) * | 2019-07-15 | 2019-11-22 | 浙江省北大信息技术高等研究院 | Object detection method, device, equipment and storage medium |
| CN110619279A (en) * | 2019-08-22 | 2019-12-27 | 天津大学 | Road traffic sign instance segmentation method based on tracking |
-
2019
- 2019-12-30 CN CN201911405434.9A patent/CN111126331A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170091950A1 (en) * | 2015-09-30 | 2017-03-30 | Fotonation Limited | Method and system for tracking an object |
| CN107316001A (en) * | 2017-05-31 | 2017-11-03 | 天津大学 | Small and intensive method for traffic sign detection in a kind of automatic Pilot scene |
| CN108320297A (en) * | 2018-03-09 | 2018-07-24 | 湖北工业大学 | A kind of video object method for real time tracking and system |
| CN110490073A (en) * | 2019-07-15 | 2019-11-22 | 浙江省北大信息技术高等研究院 | Object detection method, device, equipment and storage medium |
| CN110619279A (en) * | 2019-08-22 | 2019-12-27 | 天津大学 | Road traffic sign instance segmentation method based on tracking |
Non-Patent Citations (2)
| Title |
|---|
| BIAO HOU. ET AL: "Object detection and tracking based on convolutional neural networks for high-resolution optical remote sensing video" * |
| 唐聪等: "基于深度学习物体检测的视觉跟踪方法" * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111930877A (en) * | 2020-09-18 | 2020-11-13 | 蘑菇车联信息科技有限公司 | Map guideboard generation method and electronic equipment |
| CN111930877B (en) * | 2020-09-18 | 2021-01-05 | 蘑菇车联信息科技有限公司 | Map guideboard generation method and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108846826B (en) | Object detection method, object detection device, image processing apparatus, and storage medium | |
| CN108334881B (en) | License plate recognition method based on deep learning | |
| CN106971185B (en) | License plate positioning method and device based on full convolution network | |
| CN111967313B (en) | Unmanned aerial vehicle image annotation method assisted by deep learning target detection algorithm | |
| CN109767619B (en) | Intelligent networking pure electric vehicle running condition prediction method | |
| CN110348332B (en) | Method for extracting multi-target real-time trajectories of non-human machines in traffic video scene | |
| CN112232351B (en) | License plate recognition system based on deep neural network | |
| CN111008979A (en) | Robust night image semantic segmentation method | |
| CN113516664A (en) | Visual SLAM method based on semantic segmentation dynamic points | |
| CN111598175B (en) | Detector training optimization method based on online difficult case mining mode | |
| CN111582029A (en) | Traffic sign identification method based on dense connection and attention mechanism | |
| CN113792606A (en) | Low-cost self-supervision pedestrian re-identification model construction method based on multi-target tracking | |
| CN103605960B (en) | A kind of method for identifying traffic status merged based on different focal video image | |
| CN107247967B (en) | Vehicle window annual inspection mark detection method based on R-CNN | |
| CN111126331A (en) | Real-time guideboard detection method combining object detection and object tracking | |
| CN111354016A (en) | Unmanned aerial vehicle ship tracking method and system based on deep learning and difference value hashing | |
| CN104376538A (en) | Image sparse denoising method | |
| CN112907972A (en) | Road vehicle flow detection method and system based on unmanned aerial vehicle and computer readable storage medium | |
| CN111325811B (en) | Lane line data processing method and processing device | |
| CN111380529B (en) | Mobile device positioning method, device and system and mobile device | |
| CN112493228B (en) | Laser bird repelling method and system based on three-dimensional information estimation | |
| CN115761177A (en) | Meta-universe-oriented three-dimensional reconstruction method for cross-border financial places | |
| CN115482282A (en) | Dynamic SLAM method with multi-target tracking capability in automatic driving scene | |
| CN115512263A (en) | Dynamic visual monitoring method and device for falling object | |
| CN114445332A (en) | Multi-scale detection method based on FASTER-RCNN model |
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 |