CN111091565A - Self-adaptive motion characteristic matching and recognition bow net contact point detection method - Google Patents
Self-adaptive motion characteristic matching and recognition bow net contact point detection method Download PDFInfo
- Publication number
- CN111091565A CN111091565A CN202010000633.8A CN202010000633A CN111091565A CN 111091565 A CN111091565 A CN 111091565A CN 202010000633 A CN202010000633 A CN 202010000633A CN 111091565 A CN111091565 A CN 111091565A
- Authority
- CN
- China
- Prior art keywords
- contact point
- bow net
- pantograph
- net contact
- catenary
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/0008—Industrial image inspection checking presence/absence
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- 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
- G06T2207/10016—Video; Image sequence
-
- 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
- G06T2207/30108—Industrial image inspection
Abstract
The invention provides a pantograph-catenary contact point detection method for adaptive motion characteristic matching and identification, and relates to the technical field of railway pantograph-catenary detection and intelligent monitoring. Forming a bow net image library by a bow net image shot by a roof monitoring camera, marking out an enclosing frame taking a bow net contact point as a center, forming a bow net contact point data set by the bow net image and a marking result thereof, constructing a bow net contact point detection network, and training the detection network by using the bow net contact point data set. And matching of the pantograph and catenary contact points is completed through minimum distance judgment, a pantograph and catenary contact point information set is further established, then, whether the change frequency of the transverse speed direction of the movement of each pantograph and catenary contact point and the change range of the transverse coordinate and the longitudinal coordinate meet conditions or not is judged through analyzing the movement characteristic sequence of each pantograph and catenary contact point, and meanwhile, online learning is carried out on a detection network according to the identified real pantograph and catenary contact point data, so that the whole process can keep the adaptability to the environment.
Description
Technical Field
The invention relates to the technical field of railway pantograph-catenary detection and monitoring, mode identification and intelligent systems.
Background art:
the pantograph system is an important device of an electric locomotive traction power supply system, the pantograph contact point is an important monitoring object in pantograph operation, reflects the state of the pantograph operation, researches a pantograph contact point detection technology based on computer vision, realizes real-time and accurate monitoring, and has important significance for improving the automation and intelligence level of pantograph system detection and ensuring the safety and stability of pantograph system operation.
At present, many researches are carried out at home and abroad in the aspect of bow net non-contact detection. In 2010, a railway science research institute proposes an image processing-based method, a high-definition camera mounted on the roof of a vehicle is used for acquiring a bow net image and calculating related parameters, only one camera is used in the method, the detection structure is simple, and the precision needs to be improved. Liu etc. in 2012 installed array camera and structured light at the detection car top, carried out on-vehicle dynamic measurement based on line structured light vision measurement technique, this method has that measurement accuracy is high, reliable and stable advantage, obtains extensive application in the not high circuit of detection speed requirement simultaneously, but this method single measurement required image data volume is big, and is extremely high to image acquisition and processing requirement, has certain limitation in high-speed dynamic measurement. In 2014, Aydin I and the like detect the image edge by using a canny algorithm and extract the position of a pantograph contact point by using Hough transformation, but the algorithm cannot identify the tiny crack under the high-speed running condition and is easy to be limited by a shooting angle, part shielding and the like. In 2017, Karakose obtains edge information of a pantograph and a contact line respectively through canny edge detection, positioning and analysis of a pantograph contact point are achieved through a mode of calculating a straight line intersection point through Hough line detection, a pantograph contact surface is divided into three areas which are fault areas, danger areas and safety areas respectively, and the areas where the obtained contact points are located are judged to diagnose possible fault types. In 2018, when Shen Y and the like combine template matching with a target tracking algorithm, the detection range of a pantograph-catenary action region is narrowed in a template matching mode, a classical tracking algorithm KCF is further combined to achieve accurate tracking of a target rectangular region, finally, positioning information of pantograph-catenary contact points is obtained through calculation, and three-dimensional coordinate reconstruction and analysis are achieved through combination of binocular parameters. In 2019, Huang Z and other researches are based on contact point detection of infrared images, point detection of a pantograph and a contact net is respectively realized by utilizing two direction enhancement operators, and then positioning of a contact point is realized by adopting an improved RANSAC strategy. In 2019, Luo Y and the like propose an improved rapid RCNN for detecting pantograph faults, and positioning accuracy of candidate frames and accuracy of an algorithm are guaranteed by adjusting parameters of the rapid RCNN. In the aspect of contact point detection, the methods further need to improve the detection precision and real-time performance, and the adaptability to the change of the pantograph-catenary operating environment.
In view of this, it is necessary to innovate a bow net contact point detection method.
Disclosure of Invention
The invention aims to provide a pantograph contact point detection method for adaptive motion characteristic matching and recognition, which can effectively solve the technical problem of long-time real-time stable detection and positioning of pantograph contact points.
The purpose of the invention is realized by the following technical scheme: a bow net contact point detection method for adaptive motion feature matching and identification comprises the following steps:
1. a bow net contact point detection method for adaptive motion feature matching and recognition comprises the following steps:
step one, constructing a bow net contact point data set:
adopting a pantograph-catenary image shot by a railway locomotive roof monitoring camera to form a pantograph-catenary image library, marking an enclosing frame which takes a contact point as a center in the pantograph-catenary image in a manual marking mode, and forming a pantograph-catenary contact point data set by all pantograph-catenary images and marking results thereof;
step two, constructing and training a bow net contact point detection network:
constructing a bow net contact point detection network by adopting a currently widely used yolov3 object detection network architecture, and then training the detection network by using the bow net contact point data set constructed in the step one to enable the detection network to have the capability of initially detecting the bow net contact point;
step three, initializing the bow net contact point detection network:
setting M as the number of bow net contact points detected last time, and initially setting M to be 0; setting N as the number of currently detected bow net contact points, and initially setting N as 0;
let Q ═ LiDenotes the bow net contact point information set, where i ≦ 30,represents the motion characteristic sequence of the ith bow net contact point, wherein j is less than or equal to 300,representing the movement characteristic of the ith bow net contact point at the moment j, wherein diThe number of the bow net contact point is indicated,respectively showing the abscissa, ordinate, transverse speed, longitudinal speed, h of the bow net contact point at the moment jiIndicating the authenticity of the bow net contact point when hiWhen the contact point is 1, the bow net contact point is a real bow net contact point, and when h is equal toiWhen the contact point is equal to 0, the bow net contact point is a false bow net contact point, and h is set initiallyiWhen the value is 0, Q is empty initially;
step four, image input:
under the condition of real-time processing, extracting a video image which is acquired by a railway locomotive roof monitoring camera and stored in a pantograph-catenary image library as an input image for detecting a pantograph-catenary contact point; under the condition of off-line processing, decomposing the acquired bow net video file into an image sequence consisting of a plurality of frames, and extracting the frame images one by one as input images according to a time sequence; if the input image is empty, the whole process is stopped;
step five, bow net contact point detection:
performing contact point detection on the input image by adopting the contact point detection network obtained in the step three, and setting N as the number of currently detected contact points;
if M is equal to 0 and N is equal to 0, Q is set to be null, and the step four is skipped;
if M is 0, and N>0, setting coordinates according to the currently detected bow net contact pointThe numbers d are arranged in the order of 1 to 30iAnd is provided withj=1,hi0, is generated accordinglyAnd is added to LiThen adding LiAdding Q, setting M to be N, and jumping to the step four;
if M is greater than 0 and N is 0, setting M to be N and jumping to the step four;
if M is greater than 0 and N is greater than 0, jumping to the sixth step;
step six, bow net contact point matching:
let Ah=(xh,yh) Indicating the h bow net contact point currently detected, wherein h is less than or equal to N, xh,yhRespectively representing the abscissa and the ordinate of the bow net contact point;
calculation of AhAnd Q is LiWill have a distance to a from the coordinates of all bow net contact points at the last momenthOf minimum distanceAs AhAnd according to the matching result ofAnd AhGenerating Wherein D ofiAnd hiAndis the same, then willIs added to LiIf j +1>300, then L isiDeleting the first motion characteristic, otherwise, jumping to the fourth step;
if N is equal to M, jumping to a seventh step;
if N is present<M, then the unmatched L in QiDeleting, and then setting M to N;
if N is present>M, setting the serial numbers d of the currently unmatched N-M bow net contact pointsiSetting the coordinatesAnd is provided withj=1,hi0, is generated accordinglyAnd is added to LiThen adding LiAdding into Q, and setting M to N;
seventhly, identifying bow net contact points:
traversing the motion characteristic sequence of the pantograph contact points in the Q, if the length of the sequence is not less than 300, calculating the times of the change of the direction of the transverse speed, and if the length of the sequence is three times or more, considering the pantograph contact points corresponding to the sequence as candidate pantograph contactsContacts and sets h corresponding theretoiOtherwise, considering the pantograph contact point corresponding to the sequence as a false pantograph contact point, and setting the corresponding hi0; if the length of the sequence is less than 300, no processing is carried out;
calculating the maximum value and the minimum value of the abscissa and the ordinate in the motion characteristic sequence corresponding to all candidate bow net contact points; if the difference between the maximum value and the minimum value of the abscissa is greater than 3/4 of the image width, or the difference between the maximum value and the minimum value of the ordinate is greater than 1/4 of the input image height, the candidate contact point is considered to be also a pseudo-bow-net contact point, and h corresponding to the candidate contact point is setiIf not, the candidate pantograph contact point is considered as a true pantograph contact point, and the corresponding h is seti=1;
If the bow net contact point exists, jumping to the step eight, otherwise, jumping to the step four;
step eight, detecting network online learning:
and generating an online training set according to the data of the contact point of the real bow net, performing online learning on the detection network, and skipping to the fourth step.
Compared with the prior art, the invention has the advantages and positive effects that: the invention provides a bow net contact point detection method for adaptive motion characteristic matching and identification. The method comprises the steps of firstly constructing a pantograph contact point detection network based on a yolov3 object detection network structure, training the pantograph contact point detection network by adopting a data set obtained through manually marked pantograph images, enabling the pantograph contact point detection network to have initial pantograph contact point detection capability, detecting pantograph contact points of pantograph monitoring video images by using the detection network in a real-time detection process, judging and completing the matching of the pantograph contact points through minimum distance, further establishing a pantograph contact point information set, wherein the set comprises a motion characteristic sequence corresponding to each pantograph contact point, and identifying the truth of each pantograph contact point by analyzing the motion characteristic sequence of each pantograph contact point, so that a final pantograph contact point detection task is completed. The method of the invention utilizes the motion characteristics of the pantograph-catenary contact points to carry out motion matching on the pantograph-catenary contact points and identify the real pantograph-catenary contact points, thereby improving the accuracy of detecting the pantograph-catenary contact points, and uses an object detection network based on deep learning to position possible pantograph-catenary contact points, and improves the robustness of real-time detection of the pantograph-catenary contact points through online learning of data of the real pantograph-catenary contact points. In addition, the method can process different railway lines and locomotive conditions, and in the practical application process, accurate detection of the pantograph and catenary contact points can be realized only by modifying and enhancing the pantograph and catenary contact point data set according to specific conditions and properly configuring related parameters, so that the method has strong scene adaptability.
Drawings
FIG. 1 is a technical flow chart of the present invention
Detailed Description
The technical flow chart of the method of the invention is shown in figure 1. The method comprises the steps of firstly constructing a bow net contact point data set and a bow net contact point detection network, then training the detection network by using the data set to enable the detection network to have the initial detection capability on the bow net contact point, using the detection network to detect the bow net contact point on a bow net monitoring video image in the real-time detection process, completing the matching of the contact points through minimum distance judgment, further establishing a bow net contact point information set, wherein the set comprises a motion characteristic sequence corresponding to each bow net contact point, then identifying the true and false performance of the bow net contact point by analyzing the motion characteristic sequence of each bow net contact point, namely judging the number of times of change of the motion of the bow net contact point in the transverse speed direction and judging whether the change range of the transverse coordinate and the longitudinal coordinate meets the condition or not, thereby realizing the accurate detection on the bow net contact point, and simultaneously carrying out on-line learning on the detection network according to the identified true bow net contact point data, the method has high robustness.
Example (b):
the method can be used for different railway lines and locomotive conditions, and can accurately detect bow net contact points of various bow net monitoring video images.
Specifically, when the method is used for detecting the pantograph contact point, firstly, a pantograph image shot by a railway locomotive roof monitoring camera is used for forming a pantograph image library, then an enclosure frame which takes the contact point as the center in the pantograph image is marked in a manual marking mode, all pantograph images and marking results thereof form a contact point data set, then a contact point detection network is constructed based on a yolov3 object detection network structure, and the detection network is trained by using the contact point data set, so that the detection network has the capability of initially detecting the contact point. In the real-time detection process, the detection network is used for detecting contact points of a pantograph-catenary monitoring video image, matching of the contact points is judged and completed through the minimum distance, then a contact point information set is established, the set comprises a motion characteristic sequence corresponding to each contact point, then the number of times of change of the transverse speed direction of motion of the contact points is judged through analyzing the motion characteristic sequence of each contact point, whether the change range of the transverse coordinate and the longitudinal coordinate meets the condition or not is judged, the truth of the contact points is identified, accurate detection of the pantograph-catenary contact points is achieved, meanwhile, online learning is conducted on the detection network according to the identified true contact point data, and the whole detection process can keep the adaptability to the environment. The method can process different railway lines and locomotive conditions, and in the practical application process, accurate detection of the pantograph-catenary contact point can be realized only by modifying and enhancing the contact point data set according to specific conditions and properly configuring related parameters, so that the method has strong scene adaptability.
The method can be realized by programming in any computer programming language (such as C language), and the detection system software based on the method can realize real-time bow net contact point detection application in any PC or embedded system.
Claims (1)
1. A bow net contact point detection method for adaptive motion feature matching and recognition comprises the following steps:
step one, constructing a bow net contact point data set:
adopting a pantograph-catenary image shot by a railway locomotive roof monitoring camera to form a pantograph-catenary image library, marking an enclosing frame which takes a contact point as a center in the pantograph-catenary image in a manual marking mode, and forming a pantograph-catenary contact point data set by all pantograph-catenary images and marking results thereof;
step two, constructing and training a bow net contact point detection network:
constructing a bow net contact point detection network by adopting a currently widely used yolov3 object detection network architecture, and then training the detection network by using the bow net contact point data set constructed in the step one to enable the detection network to have the capability of initially detecting the bow net contact point;
step three, initializing the bow net contact point detection network:
setting M as the number of bow net contact points detected last time, and initially setting M to be 0; setting N as the number of currently detected bow net contact points, and initially setting N as 0;
let Q ═ LiDenotes the bow net contact point information set, where i ≦ 30,represents the motion characteristic sequence of the ith bow net contact point, wherein j is less than or equal to 300,representing the movement characteristic of the ith bow net contact point at the moment j, wherein diThe number of the bow net contact point is indicated,respectively showing the abscissa, ordinate, transverse speed, longitudinal speed, h of the bow net contact point at the moment jiIndicating the authenticity of the bow net contact point when hiWhen the contact point is 1, the bow net contact point is a real bow net contact point, and when h is equal toiWhen the contact point is equal to 0, the bow net contact point is a false bow net contact point, and h is set initiallyiWhen the value is 0, Q is empty initially;
step four, image input:
under the condition of real-time processing, extracting a video image which is acquired by a railway locomotive roof monitoring camera and stored in a pantograph-catenary image library as an input image for detecting a pantograph-catenary contact point; under the condition of off-line processing, decomposing the acquired bow net video file into an image sequence consisting of a plurality of frames, and extracting the frame images one by one as input images according to a time sequence; if the input image is empty, the whole process is stopped;
step five, bow net contact point detection:
performing contact point detection on the input image by adopting the contact point detection network obtained in the step three, and setting N as the number of currently detected contact points;
if M is equal to 0 and N is equal to 0, Q is set to be null, and the step four is skipped;
if M is 0, and N>0, setting coordinates according to the currently detected bow net contact pointThe numbers d are arranged in the order of 1 to 30iAnd is provided withj=1,hi0, is generated accordinglyAnd is added to LiThen adding LiAdding Q, setting M to be N, and jumping to the step four;
if M is greater than 0 and N is 0, setting M to be N and jumping to the step four;
if M is greater than 0 and N is greater than 0, jumping to the sixth step;
step six, bow net contact point matching:
let Ah=(xh,yh) Indicating the h bow net contact point currently detected, wherein h is less than or equal to N, xh,yhRespectively representing the abscissa and the ordinate of the bow net contact point;
calculation of AhAnd Q is LiWill have a distance to a from the coordinates of all bow net contact points at the last momenthOf minimum distanceAs AhAnd according to the matching result ofAnd AhGenerating Wherein D ofiAnd hiAndis the same, then willIs added to LiIf j +1>300, then L isiDeleting the first motion characteristic, otherwise, jumping to the fourth step;
if N is equal to M, jumping to a seventh step;
if N is present<M, then the unmatched L in QiDeleting, and then setting M to N;
if N is present>M, setting the serial numbers d of the currently unmatched N-M bow net contact pointsiSetting the coordinatesAnd is provided withj=1,hi0, is generated accordinglyAnd is added to LiThen adding LiAdding into Q, and setting M to N;
seventhly, identifying bow net contact points:
traversing the motion characteristic sequence of the pantograph contact points in the Q, if the length of the sequence is not less than 300, calculating the times of the change of the direction of the transverse speed, if the length of the sequence is three times or more, considering the pantograph contact points corresponding to the sequence as candidate pantograph contact points, and setting h corresponding to the pantograph contact pointsiOtherwise, considering the pantograph contact point corresponding to the sequence as a false pantograph contact point, and setting the corresponding hi0; if the length of the sequence is less than 300, no processing is carried out;
calculating the maximum value and the minimum value of the abscissa and the ordinate in the motion characteristic sequence corresponding to all candidate bow net contact points; if the difference between the maximum value and the minimum value of the abscissa is greater than 3/4 of the image width, or the difference between the maximum value and the minimum value of the ordinate is greater than 1/4 of the input image height, the candidate contact point is considered to be also a pseudo-bow-net contact point, and h corresponding to the candidate contact point is setiIf not, the candidate pantograph contact point is considered as a true pantograph contact point, and the corresponding h is seti=1;
If the bow net contact point exists, jumping to the step eight, otherwise, jumping to the step four;
step eight, detecting network online learning:
and generating an online training set according to the data of the contact point of the real bow net, performing online learning on the detection network, and skipping to the fourth step.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010000633.8A CN111091565B (en) | 2020-01-02 | 2020-01-02 | Self-adaptive motion characteristic matching and recognition bow net contact point detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010000633.8A CN111091565B (en) | 2020-01-02 | 2020-01-02 | Self-adaptive motion characteristic matching and recognition bow net contact point detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111091565A true CN111091565A (en) | 2020-05-01 |
CN111091565B CN111091565B (en) | 2022-02-08 |
Family
ID=70399672
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010000633.8A Active CN111091565B (en) | 2020-01-02 | 2020-01-02 | Self-adaptive motion characteristic matching and recognition bow net contact point detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111091565B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112985263A (en) * | 2021-02-09 | 2021-06-18 | 中国科学院上海微系统与信息技术研究所 | Method, device and equipment for detecting geometrical parameters of bow net |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8379921B1 (en) * | 2010-01-26 | 2013-02-19 | Verint Systems Ltd. | Method and apparatus to determine a region of interest in a video sequence based on floor detection |
CN108288055A (en) * | 2018-03-14 | 2018-07-17 | 台州智必安科技有限责任公司 | Block of bow collector of electric locomotive based on depth network and placement test and arc method for measuring |
CN207902448U (en) * | 2017-05-10 | 2018-09-25 | 中国科学院深圳先进技术研究院 | Bow net arcing Systems for optical inspection |
CN108573223A (en) * | 2018-04-03 | 2018-09-25 | 同济大学 | A kind of EMU operating environment cognitive method based on bow net video |
US20190130312A1 (en) * | 2017-10-27 | 2019-05-02 | Salesforce.Com, Inc. | Hierarchical and interpretable skill acquisition in multi-task reinforcement learning |
CN110378897A (en) * | 2019-07-25 | 2019-10-25 | 中车青岛四方机车车辆股份有限公司 | A kind of pantograph running state real-time monitoring method and device based on video |
-
2020
- 2020-01-02 CN CN202010000633.8A patent/CN111091565B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8379921B1 (en) * | 2010-01-26 | 2013-02-19 | Verint Systems Ltd. | Method and apparatus to determine a region of interest in a video sequence based on floor detection |
CN207902448U (en) * | 2017-05-10 | 2018-09-25 | 中国科学院深圳先进技术研究院 | Bow net arcing Systems for optical inspection |
US20190130312A1 (en) * | 2017-10-27 | 2019-05-02 | Salesforce.Com, Inc. | Hierarchical and interpretable skill acquisition in multi-task reinforcement learning |
CN108288055A (en) * | 2018-03-14 | 2018-07-17 | 台州智必安科技有限责任公司 | Block of bow collector of electric locomotive based on depth network and placement test and arc method for measuring |
CN108573223A (en) * | 2018-04-03 | 2018-09-25 | 同济大学 | A kind of EMU operating environment cognitive method based on bow net video |
CN110378897A (en) * | 2019-07-25 | 2019-10-25 | 中车青岛四方机车车辆股份有限公司 | A kind of pantograph running state real-time monitoring method and device based on video |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112985263A (en) * | 2021-02-09 | 2021-06-18 | 中国科学院上海微系统与信息技术研究所 | Method, device and equipment for detecting geometrical parameters of bow net |
Also Published As
Publication number | Publication date |
---|---|
CN111091565B (en) | 2022-02-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109784333B (en) | Three-dimensional target detection method and system based on point cloud weighted channel characteristics | |
CN102222341A (en) | Method and device for detecting motion characteristic point and method and device for detecting motion target | |
Zelener et al. | Cnn-based object segmentation in urban lidar with missing points | |
US20240013505A1 (en) | Method, system, medium, equipment and terminal for inland vessel identification and depth estimation for smart maritime | |
CN111797684A (en) | Binocular vision distance measuring method for moving vehicle | |
Fischer et al. | A feature descriptor for texture-less object representation using 2D and 3D cues from RGB-D data | |
Wang et al. | Autonomous landing of multi-rotors UAV with monocular gimbaled camera on moving vehicle | |
CN111091565B (en) | Self-adaptive motion characteristic matching and recognition bow net contact point detection method | |
Zhao et al. | Visual odometry-A review of approaches | |
CN106408589A (en) | Vehicle-mounted overlooking camera based vehicle movement measurement method | |
CN113393524A (en) | Target pose estimation method combining deep learning and contour point cloud reconstruction | |
Zhou et al. | BV-Net: Bin-based Vector-predicted Network for tubular solder joint detection | |
CN116665097A (en) | Self-adaptive target tracking method combining context awareness | |
Wu et al. | Adaptive ORB feature detection with a variable extraction radius in RoI for complex illumination scenes | |
CN113269118B (en) | Monocular vision forward vehicle distance detection method based on depth estimation | |
CN106056599B (en) | A kind of object recognition algorithm and device based on Object Depth data | |
Gajdošech et al. | Towards Deep Learning-based 6D Bin Pose Estimation in 3D Scans | |
CN109063543B (en) | Video vehicle weight recognition method, system and device considering local deformation | |
Yang et al. | Locator slope calculation via deep representations based on monocular vision | |
Sun et al. | The study on intelligent vehicle collision-avoidance system with vision perception and fuzzy decision making | |
Fu et al. | Vision based navigation for power transmission line inspection robot | |
Shafique et al. | Computer Vision based Autonomous Navigation in Controlled Environment | |
Gao et al. | A new method for repeated localization and matching of tunnel lining defects | |
Pan et al. | Fast vanishing point estimation based on particle swarm optimization | |
WO2024044887A1 (en) | Vision-based perception system |
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 |