CN112183255A - Underwater target visual identification and attitude estimation method based on deep learning - Google Patents

Underwater target visual identification and attitude estimation method based on deep learning Download PDF

Info

Publication number
CN112183255A
CN112183255A CN202010970281.9A CN202010970281A CN112183255A CN 112183255 A CN112183255 A CN 112183255A CN 202010970281 A CN202010970281 A CN 202010970281A CN 112183255 A CN112183255 A CN 112183255A
Authority
CN
China
Prior art keywords
network
feature
underwater
straight line
scale
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
Application number
CN202010970281.9A
Other languages
Chinese (zh)
Inventor
李乐
张辰
井新康
赵海涛
张梅洁
刘卫东
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Northwestern Polytechnical University
Original Assignee
Northwestern Polytechnical University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Northwestern Polytechnical University filed Critical Northwestern Polytechnical University
Priority to CN202010970281.9A priority Critical patent/CN112183255A/en
Publication of CN112183255A publication Critical patent/CN112183255A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • 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
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an underwater target vision recognition and posture estimation method based on deep learning. Firstly, two layers of prediction networks are added in a network structure of a basic Yolov3 algorithm, so that the rapid identification and efficient extraction of different dimensional characteristics of a target are realized; and then selecting a plurality of characteristic marks with two dimensions, the central points of which are positioned on a straight line, according to the overall structural characteristics of the target, estimating the slope of the straight line by a least square method according to the coordinate information of the characteristic marks in the target image, and further obtaining the attitude angle of the target. The underwater target vision recognition and posture estimation method based on deep learning can realize quick recognition and posture estimation of common underwater operation tools, and provides a foundation for an underwater robot mechanical arm to grab an operation tool and implement underwater operation.

Description

Underwater target visual identification and attitude estimation method based on deep learning
Technical Field
The invention belongs to the field of target recognition, and particularly relates to an underwater target recognition and attitude estimation method.
Background
A working type underwater robot is one of important equipments for exploration and development of marine resources, and is generally equipped with an underwater robot arm. The realization of the rapid and efficient recognition and attitude estimation of the underwater target is the premise of the operation of the underwater mechanical arm. Visual image processing and feature extraction are a common method for underwater target recognition and pose estimation. However, the method is limited by complex underwater environment conditions (such as dim light, turbidity and the like), the traditional underwater target visual identification method is low in identification efficiency, and the attitude estimation effect is not ideal. With the development of the deep learning theory, deep convolutional neural network sample training is an effective technical way for realizing underwater cooperative target recognition, but the effect of underwater target posture estimation is not ideal. Therefore, how to integrate the advantages of the underwater target characteristics and the deep learning method to carry out quick identification and efficient attitude estimation on the common underwater operation tool is one of the key technologies which need to be broken through when the operation type underwater robot mechanical arm works efficiently.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides an underwater target visual identification and posture estimation method based on deep learning. Firstly, two layers of prediction networks are added in a network structure of a basic Yolov3 algorithm, so that the rapid identification and efficient extraction of different dimensional characteristics of a target are realized; and then selecting a plurality of characteristic marks with two dimensions, the central points of which are positioned on a straight line, according to the overall structural characteristics of the target, estimating the slope of the straight line by a least square method according to the coordinate information of the characteristic marks in the target image, and further obtaining the attitude angle of the target. The underwater target vision recognition and posture estimation method based on deep learning can realize quick recognition and posture estimation of common underwater operation tools, and provides a foundation for an underwater robot mechanical arm to grab an operation tool and implement underwater operation.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
step 1: for each operation tool in the underwater operation tool bag, selecting N characteristic marks on the operation tool manually, wherein N is more than or equal to 3; one of the features is marked as an integral dimensional feature and is formed by the work tool itself; the rest N-1 feature marks are local dimension features and are formed by local structures or components of the working tool; the central points of the N characteristic marks are positioned on the same straight line;
step 2: work tool target identification:
step 2-1: the YOLOV3 network was modified; keeping a front-end feature extraction network of the YOLOV3 network unchanged, wherein the front-end feature extraction network outputs 5 feature graphs with the scales of 13 × 13, 26 × 26, 52 × 52, 104 × 104 and 208 × 208; two layers are added to the rear of the prediction network at the rear end of the YOLOV3 network, wherein the two layers are respectively as follows: the added first layer is a convolutional neural network, the characteristic diagram with the scale of 52 x 52 output by the front-end feature extraction network is up-sampled to obtain a characteristic diagram of 104 x 104, and then the characteristic diagram is fused with the characteristic diagram with the scale of 104 x 104 output by the front-end feature extraction network; the second layer added is a convolutional neural network, the characteristic diagram with the scale of 104 × 104 output by the front-end feature extraction network is up-sampled to obtain a characteristic diagram with the scale of 208 × 208, and then the characteristic diagram is fused with the characteristic diagram with the scale of 208 × 208 output by the front-end feature extraction network; thereby generating a modified YOLOV3 network;
step 2-2, collecting A images of the underwater operation tool as a training set, processing the training set by adopting a k-means clustering algorithm to obtain anchor frames on a plurality of scales, training the improved YOLOV3 network obtained in the step 2-1, and training iteration steps are B steps; obtaining a trained improved YOLOV3 network;
step 2-3: inputting the underwater operation tool image into the improved YOLOV3 network trained in the step 2-2, extracting the operation tool image characteristics through the front-end characteristic extraction network of the improved YOLOV3 network trained to obtain 5-scale characteristic graphs, wherein the 5 scales are 13 × 13, 26 × 26, 52 × 52, 104 × 104 and 208 × 208 respectively;
step 2-4: detecting and fusing the feature maps of 5 scales obtained in the step 2-3 by utilizing the trained rear-end prediction network of the improved YOLOV3 network to obtain the coordinate (t) of the center point of the bounding box corresponding to any feature mark of the manually marked working tool in the step 1 on any scalex,ty) Width twAnd height th；
Step 2-5: coordinate (t) of the center point of the boundary frame obtained in the step 2-4x,ty) Width twAnd height thThe normalization processing is performed by using the following formulas (1) to (4), specifically as follows:
bx＝σ(tx)+cx (1)
by＝σ(ty)+cy (2)
Figure BDA0002683806470000021
Figure BDA0002683806470000022
wherein: sigma is a sigmoid function, and the value range is (0, 1); (b)x,by) As coordinates of the center point of the bounding box after normalization, bwFor the width of the bounding box after normalization, bhTo the normalized bounding box height, cx,cyValues are all 1; p is a radical ofw,phMapping width and height in the feature map for the anchor box on the corresponding scale;
step 2-6: repeating the steps 2-4 and 2-5, obtaining the coordinates, widths and heights of the central points of all the bounding boxes corresponding to the N feature labels on 5 scales through the rear-end prediction network of the trained improved YOLOV3 network, and then carrying out normalization processing;
step 2-7: classifying the coordinates, widths and heights of the central point of the boundary frame after the normalization processing in the steps 2-6 according to different feature marks to perform non-maximum suppression processing to obtain a final recognition and feature extraction result, namely the coordinates (x) of the central points of the boundary frame of N feature marks in the working tool1,y1)、(x2,y2)、…(xN,yN) Width w of bounding box1,w2,...,wNAnd height h1,h2,...,hN；
And step 3: using least square method to the N central points (x)1,y1)、(x2,y2)、…(xN,yN) Fitting to obtain a straight line, and estimating the slope k of the straight line through the following formula (5);
let the equation of a straight line be: y — kx + b, the slope k of the line is estimated as:
Figure BDA0002683806470000031
wherein
Figure BDA0002683806470000032
Is x1,x2,…,xNAverage value of (x)i，yi) The coordinate of the center point of the ith bounding box is shown, and b is a straight line intercept;
and 4, step 4: and (4) calculating the attitude angle of the underwater operation tool in the image plane according to the slope k of the straight line obtained in the step (3).
Preferably, N is 3.
Preferably, a is 500 and B is 10000.
The invention has the beneficial effects that:
1. compared with the traditional underwater target visual identification method, the underwater target identification algorithm of the improved YOLOV3 network can realize quick identification and effective extraction of multi-dimensional characteristic information of the underwater target, and improves the accuracy and the rapidity of underwater target identification.
2. The underwater target attitude estimation algorithm fusing the multi-dimensional feature markers can utilize coordinate information of a plurality of underwater target feature markers with central points located on a straight line and belonging to two dimensions in a target image, and estimates the slope of the straight line and obtains the attitude angle of the target by a least square method.
Drawings
Fig. 1 is a schematic diagram of a modified YOLOV3 network structure according to the present invention.
Fig. 2 is a drawing for selecting a wrench feature mark according to an embodiment of the present invention.
Fig. 3 is a diagram of a feature label recognition result according to an embodiment of the present invention.
In the figure: 1-wrench body, 2-wrench head, 3-wrench tail, 4-improved YOLOV3 network, 5-front-end feature extraction network, 6-rear-end prediction network, 7-wrench tail feature label boundary box, 8-wrench head feature label boundary box, and 9-wrench body feature label boundary box.
Detailed Description
The invention is further illustrated with reference to the following figures and examples.
As shown in fig. 1, the invention provides a method for underwater target visual recognition and attitude estimation based on deep learning, which comprises the following steps:
step 1: for each operation tool in the underwater operation tool bag, selecting N characteristic marks on the operation tool manually, wherein N is more than or equal to 3; one of the features is marked as an integral dimensional feature and is formed by the work tool itself; the rest N-1 feature marks are local dimension features and are formed by local structures or components of the working tool; the central points of the N characteristic marks are positioned on the same straight line;
step 2: work tool target identification:
step 2-1: the YOLOV3 network was modified; keeping a front-end feature extraction network of the YOLOV3 network unchanged, wherein the front-end feature extraction network outputs 5 feature graphs with the scales of 13 × 13, 26 × 26, 52 × 52, 104 × 104 and 208 × 208; two layers are added to the rear of the prediction network at the rear end of the YOLOV3 network, wherein the two layers are respectively as follows: the added first layer is a convolutional neural network, the characteristic diagram with the scale of 52 x 52 output by the front-end feature extraction network is up-sampled to obtain a characteristic diagram of 104 x 104, and then the characteristic diagram is fused with the characteristic diagram with the scale of 104 x 104 output by the front-end feature extraction network; the second layer added is a convolutional neural network, the characteristic diagram with the scale of 104 × 104 output by the front-end feature extraction network is up-sampled to obtain a characteristic diagram with the scale of 208 × 208, and then the characteristic diagram is fused with the characteristic diagram with the scale of 208 × 208 output by the front-end feature extraction network; thereby generating a modified YOLOV3 network;
step 2-2, collecting 500 images of the underwater operation tool in total number as a training set, processing the training set by adopting a k-means clustering algorithm to obtain anchor frames on a plurality of scales, training the improved Yolov3 network obtained in the step 2-1, and training the iteration steps to 10000 steps; obtaining a trained improved YOLOV3 network;
step 2-3: inputting the underwater operation tool image into the improved YOLOV3 network trained in the step 2-2, extracting the operation tool image characteristics through the front-end characteristic extraction network of the improved YOLOV3 network trained to obtain 5-scale characteristic graphs, wherein the 5 scales are 13 × 13, 26 × 26, 52 × 52, 104 × 104 and 208 × 208 respectively;
step 2-4: detecting and fusing the feature maps of 5 scales obtained in the step 2-3 by utilizing the trained rear-end prediction network of the improved YOLOV3 network to obtain the coordinate (t) of the center point of the bounding box corresponding to any feature mark of the manually marked working tool in the step 1 on any scalex,ty) Width twAnd height th；
Step 2-5: coordinate (t) of the center point of the boundary frame obtained in the step 2-4x,ty) Width twAnd height thThe normalization processing is performed by using the following formulas (1) to (4), specifically as follows:
bx＝σ(tx)+cx (1)
by＝σ(ty)+cy (2)
Figure BDA0002683806470000051
Figure BDA0002683806470000052
wherein: sigma is a sigmoid function, and the value range is (0, 1); (b)x,by) As coordinates of the center point of the bounding box after normalization, bwFor the width of the bounding box after normalization, bhTo the normalized bounding box height, cx,cyValues are all 1; p is a radical ofw,phMapping width and height in the feature map for the anchor box on the corresponding scale;
step 2-6: repeating the steps 2-4 and 2-5, obtaining the coordinates, widths and heights of the central points of all the bounding boxes corresponding to the N feature labels on 5 scales through the rear-end prediction network of the trained improved YOLOV3 network, and then carrying out normalization processing;
step 2-7: classifying the coordinates, widths and heights of the central point of the boundary frame after the normalization processing in the steps 2-6 according to different feature marks to perform non-maximum suppression processing to obtain a final recognition and feature extraction result, namely the coordinates (x) of the central points of the boundary frame of N feature marks in the working tool1,y1)、(x2,y2)、…(xN,yN) Width w of bounding box1,w2,...,wNAnd height h1,h2,...,hN；
And step 3: using least square method to the N central points (x)1,y1)、(x2,y2)、…(xN,yN) Fitting to obtain a straight line, and estimating the slope k of the straight line through the following formula (5);
let the equation of a straight line be: y — kx + b, the slope k of the line is estimated as:
Figure BDA0002683806470000053
wherein
Figure BDA0002683806470000054
Is x1,x2,…,xNAverage value of (x)i，yi) The coordinate of the center point of the ith bounding box is shown, and b is a straight line intercept;
and 4, step 4: and (4) calculating the attitude angle of the underwater operation tool in the image plane according to the slope k of the straight line obtained in the step (3).
The specific embodiment is as follows:
1. according to the overall structure of the working tool, three feature marks with central points located on the same straight line and belonging to two dimensions are selected. Wherein the first feature is labeled as the work tool itself (global dimension), the second feature and the third feature are labeled as local structures or components on the work tool (local dimension). For a common tool wrench, as shown in fig. 2, three feature labels are respectively selected according to the special mechanical structure of the wrench: a wrench body 1, a wrench head 2, and a wrench tail 3.
2. Inputting the underwater operation tool image into an underwater target recognition algorithm of improved YOLOV3 to obtain three bounding boxes corresponding to the three feature labels and coordinates (x) of center points of the three bounding boxes1,y1)、(x2,y2) And (x)3,y3)。
3. Using least squares method to pair (x)1,y1)、(x2,y2) And (x)3,y3) And fitting the three central points to obtain a straight line, and estimating the slope k of the straight line by using a formula (5).
4. In an image plane coordinate system, the true south direction is taken as 0 degree direction, the angle range is clockwise 0-360 degrees, the attitude angle of the underwater target in the image plane is obtained by combining the relation between the slope k of the straight line and the inclination angle according to the regulation of the angle direction, and a foundation is provided for the underwater robot mechanical arm to grab a working tool and implement underwater operation.

Claims (3)

1. An underwater target vision recognition and posture estimation method based on deep learning is characterized by comprising the following steps:
step 1: for each operation tool in the underwater operation tool bag, selecting N characteristic marks on the operation tool manually, wherein N is more than or equal to 3; one of the features is marked as an integral dimensional feature and is formed by the work tool itself; the rest N-1 feature marks are local dimension features and are formed by local structures or components of the working tool; the central points of the N characteristic marks are positioned on the same straight line;
step 2: work tool target identification:
step 2-1: the YOLOV3 network was modified; keeping a front-end feature extraction network of the YOLOV3 network unchanged, wherein the front-end feature extraction network outputs 5 feature graphs with the scales of 13 × 13, 26 × 26, 52 × 52, 104 × 104 and 208 × 208; two layers are added to the rear of the prediction network at the rear end of the YOLOV3 network, wherein the two layers are respectively as follows: the added first layer is a convolutional neural network, the characteristic diagram with the scale of 52 x 52 output by the front-end feature extraction network is up-sampled to obtain a characteristic diagram of 104 x 104, and then the characteristic diagram is fused with the characteristic diagram with the scale of 104 x 104 output by the front-end feature extraction network; the second layer added is a convolutional neural network, the characteristic diagram with the scale of 104 × 104 output by the front-end feature extraction network is up-sampled to obtain a characteristic diagram with the scale of 208 × 208, and then the characteristic diagram is fused with the characteristic diagram with the scale of 208 × 208 output by the front-end feature extraction network; thereby generating a modified YOLOV3 network;
step 2-2, collecting A images of the underwater operation tool as a training set, processing the training set by adopting a k-means clustering algorithm to obtain anchor frames on a plurality of scales, training the improved YOLOV3 network obtained in the step 2-1, and training iteration steps are B steps; obtaining a trained improved YOLOV3 network;
step 2-3: inputting the underwater operation tool image into the improved YOLOV3 network trained in the step 2-2, extracting the operation tool image characteristics through the front-end characteristic extraction network of the improved YOLOV3 network trained to obtain 5-scale characteristic graphs, wherein the 5 scales are 13 × 13, 26 × 26, 52 × 52, 104 × 104 and 208 × 208 respectively;
step 2-4: detecting and fusing the feature maps of 5 scales obtained in the step 2-3 by utilizing the trained rear-end prediction network of the improved YOLOV3 network to obtain the coordinate (t) of the center point of the bounding box corresponding to any feature mark of the manually marked working tool in the step 1 on any scalex,ty) Width twAnd height th；
Step 2-5: coordinate (t) of the center point of the boundary frame obtained in the step 2-4x,ty) Width twAnd height thThe normalization processing is performed by using the following formulas (1) to (4), specifically as follows:
bx＝σ(tx)+cx (1)
by＝σ(ty)+cy (2)
Figure FDA0002683806460000021
Figure FDA0002683806460000022
wherein: sigma is a sigmoid function, and the value range is (0, 1); (b)x,by) As coordinates of the center point of the bounding box after normalization, bwFor the width of the bounding box after normalization, bhTo the normalized bounding box height, cx,cyValues are all 1; p is a radical ofw,phMapping width and height in the feature map for the anchor box on the corresponding scale;
step 2-6: repeating the steps 2-4 and 2-5, obtaining the coordinates, widths and heights of the central points of all the bounding boxes corresponding to the N feature labels on 5 scales through the rear-end prediction network of the trained improved YOLOV3 network, and then carrying out normalization processing;
step 2-7: classifying the coordinates, widths and heights of the central point of the boundary frame after the normalization processing in the steps 2-6 according to different feature marks to perform non-maximum suppression processing to obtain a final recognition and feature extraction result, namely the coordinates (x) of the central points of the boundary frame of N feature marks in the working tool1,y1)、(x2,y2)、…(xN,yN) Width w of bounding box1,w2,...,wNAnd height h1,h2,...,hN；
And step 3: using least square method to the N central points (x)1,y1)、(x2,y2)、…(xN,yN) Fitting to obtain a straight line, and estimating the slope k of the straight line through the following formula (5);
let the equation of a straight line be: y — kx + b, the slope k of the line is estimated as:
Figure FDA0002683806460000023
wherein
Figure FDA0002683806460000024
Is x1,x2,…,xNAverage value of (x)i，yi) The coordinate of the center point of the ith bounding box is shown, and b is a straight line intercept;
and 4, step 4: and (4) calculating the attitude angle of the underwater operation tool in the image plane according to the slope k of the straight line obtained in the step (3).
2. The deep learning-based underwater target vision recognition and pose estimation method according to claim 1, wherein N-3.
3. The deep learning-based underwater target vision recognition and pose estimation method according to claim 1, wherein a is 500 and B is 10000.
CN202010970281.9A 2020-09-15 2020-09-15 Underwater target visual identification and attitude estimation method based on deep learning Pending CN112183255A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010970281.9A CN112183255A (en) 2020-09-15 2020-09-15 Underwater target visual identification and attitude estimation method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010970281.9A CN112183255A (en) 2020-09-15 2020-09-15 Underwater target visual identification and attitude estimation method based on deep learning

Publications (1)

Publication Number Publication Date
CN112183255A true CN112183255A (en) 2021-01-05

Family

ID=73921185

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010970281.9A Pending CN112183255A (en) 2020-09-15 2020-09-15 Underwater target visual identification and attitude estimation method based on deep learning

Country Status (1)

Country Link
CN (1) CN112183255A (en)

Citations (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104318233A (en) * 2014-10-19 2015-01-28 温州大学 Method for horizontal tilt correction of number plate image
CN105678296A (en) * 2015-12-30 2016-06-15 小米科技有限责任公司 Method and apparatus for determining angle of inclination of characters
CN107590498A (en) * 2017-09-27 2018-01-16 哈尔滨工业大学 A kind of self-adapted car instrument detecting method based on Character segmentation level di- grader
CN107609453A (en) * 2016-07-11 2018-01-19 北京君正集成电路股份有限公司 A kind of license plate image correction, registration number character dividing method and equipment
KR101885839B1 (en) * 2017-03-14 2018-08-06 중앙대학교 산학협력단 System and Method for Key point Selecting for Object Tracking
CN108764178A (en) * 2018-05-31 2018-11-06 中国民航大学 A kind of three modal characteristics image posture synchronous acquisition device of finger and control method
CN108961221A (en) * 2018-06-15 2018-12-07 哈尔滨工业大学 A kind of aviation plug situ static image detection algorithm
CN109508710A (en) * 2018-10-23 2019-03-22 东华大学 Based on the unmanned vehicle night-environment cognitive method for improving YOLOv3 network
CN110210320A (en) * 2019-05-07 2019-09-06 南京理工大学 The unmarked Attitude estimation method of multiple target based on depth convolutional neural networks
CN110458798A (en) * 2019-06-20 2019-11-15 长沙理工大学 Damper defective vision detection method, system and medium based on critical point detection
CN110472573A (en) * 2019-08-14 2019-11-19 北京思图场景数据科技服务有限公司 A kind of human body behavior analysis method, equipment and computer storage medium based on body key point
CN110796168A (en) * 2019-09-26 2020-02-14 江苏大学 Improved YOLOv 3-based vehicle detection method
CN110909738A (en) * 2019-11-15 2020-03-24 杭州远鉴信息科技有限公司 Automatic reading method of pointer instrument based on key point detection
CN111104940A (en) * 2018-10-26 2020-05-05 深圳怡化电脑股份有限公司 Image rotation correction method and device, electronic equipment and storage medium
CN111353544A (en) * 2020-03-05 2020-06-30 天津城建大学 Improved Mixed Pooling-Yolov 3-based target detection method
CN111401148A (en) * 2020-02-27 2020-07-10 江苏大学 Road multi-target detection method based on improved multilevel YO L Ov3
CN111488874A (en) * 2020-04-03 2020-08-04 中国农业大学 Method and system for correcting inclination of pointer instrument
CN111553406A (en) * 2020-04-24 2020-08-18 上海锘科智能科技有限公司 Target detection system, method and terminal based on improved YOLO-V3

Patent Citations (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104318233A (en) * 2014-10-19 2015-01-28 温州大学 Method for horizontal tilt correction of number plate image
CN105678296A (en) * 2015-12-30 2016-06-15 小米科技有限责任公司 Method and apparatus for determining angle of inclination of characters
CN107609453A (en) * 2016-07-11 2018-01-19 北京君正集成电路股份有限公司 A kind of license plate image correction, registration number character dividing method and equipment
KR101885839B1 (en) * 2017-03-14 2018-08-06 중앙대학교 산학협력단 System and Method for Key point Selecting for Object Tracking
CN107590498A (en) * 2017-09-27 2018-01-16 哈尔滨工业大学 A kind of self-adapted car instrument detecting method based on Character segmentation level di- grader
CN108764178A (en) * 2018-05-31 2018-11-06 中国民航大学 A kind of three modal characteristics image posture synchronous acquisition device of finger and control method
CN108961221A (en) * 2018-06-15 2018-12-07 哈尔滨工业大学 A kind of aviation plug situ static image detection algorithm
CN109508710A (en) * 2018-10-23 2019-03-22 东华大学 Based on the unmanned vehicle night-environment cognitive method for improving YOLOv3 network
CN111104940A (en) * 2018-10-26 2020-05-05 深圳怡化电脑股份有限公司 Image rotation correction method and device, electronic equipment and storage medium
CN110210320A (en) * 2019-05-07 2019-09-06 南京理工大学 The unmarked Attitude estimation method of multiple target based on depth convolutional neural networks
CN110458798A (en) * 2019-06-20 2019-11-15 长沙理工大学 Damper defective vision detection method, system and medium based on critical point detection
CN110472573A (en) * 2019-08-14 2019-11-19 北京思图场景数据科技服务有限公司 A kind of human body behavior analysis method, equipment and computer storage medium based on body key point
CN110796168A (en) * 2019-09-26 2020-02-14 江苏大学 Improved YOLOv 3-based vehicle detection method
CN110909738A (en) * 2019-11-15 2020-03-24 杭州远鉴信息科技有限公司 Automatic reading method of pointer instrument based on key point detection
CN111401148A (en) * 2020-02-27 2020-07-10 江苏大学 Road multi-target detection method based on improved multilevel YO L Ov3
CN111353544A (en) * 2020-03-05 2020-06-30 天津城建大学 Improved Mixed Pooling-Yolov 3-based target detection method
CN111488874A (en) * 2020-04-03 2020-08-04 中国农业大学 Method and system for correcting inclination of pointer instrument
CN111553406A (en) * 2020-04-24 2020-08-18 上海锘科智能科技有限公司 Target detection system, method and terminal based on improved YOLO-V3

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
CHANGYOU LI ET AL: "Skew detection of track images based on wavelet transform and linear least square fitting", 《2009 INTERNATIONAL CONFERENCE ON INFORMATION AND AUTOMATION》 *
范丽等: "基于YOLOv3模型的实时行人检测改进算法", 《山西大学学报（自然科学版）》 *
陆亚涵: "变电站电力表计检测与读数识别算法研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 *

Similar Documents

Publication Publication Date Title
CN111775152B (en) Method and system for guiding mechanical arm to grab scattered stacked workpieces based on three-dimensional measurement
CN108665496B (en) End-to-end semantic instant positioning and mapping method based on deep learning
CN111080693A (en) Robot autonomous classification grabbing method based on YOLOv3
WO2022017131A1 (en) Point cloud data processing method and device, and intelligent driving control method and device
CN112509063A (en) Mechanical arm grabbing system and method based on edge feature matching
CN111862201A (en) Deep learning-based spatial non-cooperative target relative pose estimation method
CN109048918B (en) Visual guide method for wheelchair mechanical arm robot
CN103413352A (en) Scene three-dimensional reconstruction method based on RGBD multi-sensor fusion
CN110533716B (en) Semantic SLAM system and method based on 3D constraint
CN112734844B (en) Monocular 6D pose estimation method based on octahedron
CN110135277B (en) Human behavior recognition method based on convolutional neural network
Arain et al. Improving underwater obstacle detection using semantic image segmentation
CN113888461A (en) Method, system and equipment for detecting defects of hardware parts based on deep learning
CN113034600A (en) Non-texture planar structure industrial part identification and 6D pose estimation method based on template matching
CN114140418A (en) Seven-degree-of-freedom grabbing posture detection method based on RGB image and depth image
CN111914832B (en) SLAM method of RGB-D camera under dynamic scene
Sun et al. Accurate lane detection with atrous convolution and spatial pyramid pooling for autonomous driving
CN113343875A (en) Driving region sensing method for robot
CN116030130A (en) Hybrid semantic SLAM method in dynamic environment
CN113295171B (en) Monocular vision-based attitude estimation method for rotating rigid body spacecraft
CN106650814B (en) Outdoor road self-adaptive classifier generation method based on vehicle-mounted monocular vision
CN112183255A (en) Underwater target visual identification and attitude estimation method based on deep learning
CN107248171B (en) Triangulation-based monocular vision odometer scale recovery method
CN113554705A (en) Robust positioning method for laser radar in changing scene
CN113592947A (en) Visual odometer implementation method of semi-direct method

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
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20210105