CN110084777A - A kind of micro parts positioning and tracing method based on deep learning - Google Patents
A kind of micro parts positioning and tracing method based on deep learning Download PDFInfo
- Publication number
- CN110084777A CN110084777A CN201811307038.8A CN201811307038A CN110084777A CN 110084777 A CN110084777 A CN 110084777A CN 201811307038 A CN201811307038 A CN 201811307038A CN 110084777 A CN110084777 A CN 110084777A
- Authority
- CN
- China
- Prior art keywords
- target
- micro parts
- network
- deep learning
- tracing method
- 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
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/223—Analysis of motion using block-matching
- G06T7/231—Analysis of motion using block-matching using full search
-
- 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
- G06T7/251—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving models
-
- 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
- G06T2207/20—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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
A kind of micro parts positioning and tracing method based on deep learning, traditional correlation filtering method is using Gradient Features and color characteristic, target tracking effect in the case where rapid deformation or discoloration is caused to be deteriorated, a kind of micro parts positioning and tracing method based on deep learning is proposed thus, the method is using the multiple dimensioned depth characteristic obtained by FPN network, it can be according to the size selection feature of target, and combine template matching network, greatly improve the recognition capability of micro parts, to brightness change, the influence factors such as image blur have stronger adaptability;In addition two kinds of negative samples pair are also added into, chaff interferent can be effectively distinguished, increase the robustness of model;It is eventually adding verifying network and full figure search module, its position is quickly positioned after target loss, entire model is obtained by off-line form training, it is not necessary to more new model, so still having the effect of real-time tracking under the premise of keeping high performance.
Description
Technical field
The invention mainly relates to artificial intelligence application fields and micro assemby field.More particularly to a kind of based on deep learning
Micro parts positioning and tracing method.
Background technique
Micro parts tracking be it is a kind of during micro assemby to a kind of technology of target identification, positioning and tracking,
In recent years, demand of the people to manufacturing industry product is increasing, and the requirement to product quality is also higher and higher, same with this
When, the accurate microminiature electronic product obtained by micro assemby is with compact-sized, performance is stable, low energy consumption, anti-interference ability
The features such as strong, is used widely in every field, since there are feeding, blanking, components to move during micro assemby, fixed
Target following technology introducing micro-vision field is of great significance by the links such as position.
Micro parts are generally rigid objects, so generalling use region or signature tracking algorithm.Signature tracking algorithm
Common feature has color, gradient, edge, texture etc., for rapid deformation, discoloration, fuzzy target following effect and pays no attention to
Think, during micro assemby, micro-vision field range is limited, and a minute movement may be such that part loses from the visual field
It loses, simultaneously because the depth of field of microlens is smaller, target part imaging occurs during the variation of operating distance will lead to tracking
Different degrees of fuzzy and change in size situation, if the complexity for carrying out feature combination, and algorithm being made to become, it is difficult to guarantee real
Shi Xing, feature class track algorithm are faced with stern challenge.
To sum up, to overcome the difficulties such as visual field loss, image blur, a kind of region class track algorithm is proposed, it is used
Multiple dimensioned depth characteristic can effectively solve to track failure problem caused by because of image blur or change in size, and verifying network
With global search module, target position can be positioned rapidly in the case where visual field loss.It is transported parallel using convolution operation and GPU
Calculation can greatly improve calculating speed.
Summary of the invention
The technical problem to be solved in the present invention is to provide a kind of positioning and tracing method of micro parts based on deep learning,
It is characterized in that the positioning and tracing method of the micro parts includes the following steps.
A, micro parts training set is made.
B, object matching network is constructed.
C, RPN network extracts object candidate area.
D, training network, obtains final micro parts trace model.
E, judge whether target loses using verifying network.
If F, target is lost, full figure search module is opened.
G, label target position.
Preferably, micro parts training set is made in the step A, using two kinds of data sets of video and image.Video is every
It was once sampled every 6 seconds, using the target in first frame as template, other frames are as region of search;Image data set includes just
Sample is to the negative sample of, the same category to, different classes of negative sample pair.
Preferably, the step B building object matching network includes the following steps.
A, mobile terminal network model MobileNet is selected.
B, the Fusion Features from deep layer to shallow-layer are carried out to MobileNet network using FPN algorithm.
C, the characteristic pattern for obtaining template and region of search carries out relevant operation.
Preferably, judge whether target loses using verifying network in the step E to include the following steps.
A, trained VGG16 model is selected.
B, center loss function is added.
C, it is trained with the micro parts data set collected.
Preferably, if it is when verifying network to judge that the object of tracking is not target, from picture that target, which is lost, in the step F
The upper left corner starts, and determines the local search area of four times of sizes of a target, successively searches for full figure, and lateral step-length is that target is long
Half, longitudinal step-length are the wide half of target.
The beneficial effects of the invention are as follows.
(1) due to the multiple dimensioned depth characteristic obtained using FPN network, so in feelings such as objective fuzzy, change in size
It remains to accurately track target under condition.
(2), since verifying network and full figure search module is added, it can judge in time whether target loses, and losing target
Quick position location afterwards.
(3) increased three kinds of image training samples pair, can effectively distinguish chaff interferent, increase the robustness of model.
Detailed description of the invention:
Fig. 1 is the flow chart of the tracking of micro parts.
Fig. 2 is FPN network characterization fusion figure.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, referring to attached drawing, to of the invention further detailed
It describes in detail bright.
Step S1: data collection is carried out using microscope and industrial CCD camera, and makes training set.
Step S2: matching network building is carried out using MobileNet network and FPN network.
Step S3: the candidate region of target is obtained by RPN network, and non-maxima suppression is carried out to candidate region.
Step S4: it is trained using training the set pair analysis model.
Step S5: judging to track whether object is target using verifying network, if not opening whole district's search.
Step S6: label target position.
Claims (5)
1. the micro parts positioning and tracing method based on deep learning, it is characterised in that the micro parts tracking include with
Lower step:
Make micro parts training set;
Construct object matching network;
RPN network extracts object candidate area, and carries out non-maxima suppression to candidate region;
Training network, obtains final micro parts trace model;
Judge whether target loses using verifying network;
If target is lost, full figure search module is opened;
Label target position.
2. the micro parts positioning and tracing method according to claim 1 based on deep learning, which is characterized in that step
Micro parts training set is made in A, using two kinds of data sets of video and image;Video was once sampled every 6 seconds, with first frame
In target as template, other frames are as region of search;Image data set includes positive sample to the negative sample of, the same category
To, different classes of negative sample pair.
3. the micro parts positioning and tracing method according to claim 1 based on deep learning, which is characterized in that step
B construct object matching network the following steps are included:
Select mobile terminal network model MobileNet;
The Fusion Features from deep layer to shallow-layer are carried out to MobileNet network using FPN algorithm;
The characteristic pattern that template and region of search are obtained carries out relevant operation.
4. the micro parts positioning and tracing method according to claim 1 based on deep learning, which is characterized in that step
In E using verifying network judge target whether lose the following steps are included:
Select trained VGG16 model;
Addition center loss function;
It is trained with the micro parts data set collected.
5. the micro parts positioning and tracing method according to claim 1 based on deep learning, which is characterized in that step
If it is when verifying network to judge that the object of tracking is not target, since the picture upper left corner, to determine a target that target, which is lost, in F
The local search area of four times of sizes successively searches for full figure, and lateral step-length is the long half of target, and longitudinal step-length is that target is wide
Half.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811307038.8A CN110084777A (en) | 2018-11-05 | 2018-11-05 | A kind of micro parts positioning and tracing method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811307038.8A CN110084777A (en) | 2018-11-05 | 2018-11-05 | A kind of micro parts positioning and tracing method based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110084777A true CN110084777A (en) | 2019-08-02 |
Family
ID=67412882
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811307038.8A Pending CN110084777A (en) | 2018-11-05 | 2018-11-05 | A kind of micro parts positioning and tracing method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110084777A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111340850A (en) * | 2020-03-20 | 2020-06-26 | 军事科学院***工程研究院***总体研究所 | Ground target tracking method of unmanned aerial vehicle based on twin network and central logic loss |
CN112566016A (en) * | 2020-11-19 | 2021-03-26 | 西安理工大学 | Deep learning and block chain based maintenance tool LoRa positioning method |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106709936A (en) * | 2016-12-14 | 2017-05-24 | 北京工业大学 | Single target tracking method based on convolution neural network |
CN107066990A (en) * | 2017-05-04 | 2017-08-18 | 厦门美图之家科技有限公司 | A kind of method for tracking target and mobile device |
CN107564034A (en) * | 2017-07-27 | 2018-01-09 | 华南理工大学 | The pedestrian detection and tracking of multiple target in a kind of monitor video |
CN107679250A (en) * | 2017-11-01 | 2018-02-09 | 浙江工业大学 | A kind of multitask layered image search method based on depth own coding convolutional neural networks |
US20180165551A1 (en) * | 2016-12-08 | 2018-06-14 | Intel Corporation | Technologies for improved object detection accuracy with multi-scale representation and training |
CN108647668A (en) * | 2018-05-21 | 2018-10-12 | 北京亮亮视野科技有限公司 | The construction method of multiple dimensioned lightweight Face datection model and the method for detecting human face based on the model |
CN108665483A (en) * | 2018-05-04 | 2018-10-16 | 浙江工业大学 | A kind of cancer cell tracking based on multi-feature fusion |
-
2018
- 2018-11-05 CN CN201811307038.8A patent/CN110084777A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180165551A1 (en) * | 2016-12-08 | 2018-06-14 | Intel Corporation | Technologies for improved object detection accuracy with multi-scale representation and training |
CN106709936A (en) * | 2016-12-14 | 2017-05-24 | 北京工业大学 | Single target tracking method based on convolution neural network |
CN107066990A (en) * | 2017-05-04 | 2017-08-18 | 厦门美图之家科技有限公司 | A kind of method for tracking target and mobile device |
CN107564034A (en) * | 2017-07-27 | 2018-01-09 | 华南理工大学 | The pedestrian detection and tracking of multiple target in a kind of monitor video |
CN107679250A (en) * | 2017-11-01 | 2018-02-09 | 浙江工业大学 | A kind of multitask layered image search method based on depth own coding convolutional neural networks |
CN108665483A (en) * | 2018-05-04 | 2018-10-16 | 浙江工业大学 | A kind of cancer cell tracking based on multi-feature fusion |
CN108647668A (en) * | 2018-05-21 | 2018-10-12 | 北京亮亮视野科技有限公司 | The construction method of multiple dimensioned lightweight Face datection model and the method for detecting human face based on the model |
Non-Patent Citations (2)
Title |
---|
吴叶兰 等: "基于显微视觉的微小型零件跟踪***", 《传感器与微***》 * |
陈慧岩: "《智能车辆理论与应用》", 31 July 2018 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111340850A (en) * | 2020-03-20 | 2020-06-26 | 军事科学院***工程研究院***总体研究所 | Ground target tracking method of unmanned aerial vehicle based on twin network and central logic loss |
CN112566016A (en) * | 2020-11-19 | 2021-03-26 | 西安理工大学 | Deep learning and block chain based maintenance tool LoRa positioning method |
CN112566016B (en) * | 2020-11-19 | 2023-08-01 | 上海九乾航空科技有限公司 | Deep learning and blockchain-based method for positioning loRa (loRa) of maintenance tool |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111223088B (en) | Casting surface defect identification method based on deep convolutional neural network | |
CN110443827B (en) | Unmanned aerial vehicle video single-target long-term tracking method based on improved twin network | |
CN104749184B (en) | Automatic optical detection method and system | |
CN105631880B (en) | Lane line dividing method and device | |
CN109285179A (en) | A kind of motion target tracking method based on multi-feature fusion | |
CN107808376B (en) | Hand raising detection method based on deep learning | |
CN106845408A (en) | A kind of street refuse recognition methods under complex environment | |
CN107403175A (en) | Visual tracking method and Visual Tracking System under a kind of movement background | |
CN103955949B (en) | Moving target detecting method based on Mean-shift algorithm | |
CN108198221A (en) | A kind of automatic stage light tracking system and method based on limb action | |
CN109045676B (en) | Chinese chess recognition learning algorithm and robot intelligent system and method based on algorithm | |
CN107886131A (en) | One kind is based on convolutional neural networks detection circuit board element polarity method and apparatus | |
CN111914761A (en) | Thermal infrared face recognition method and system | |
CN109166137A (en) | For shake Moving Object in Video Sequences detection algorithm | |
CN110765814A (en) | Blackboard writing behavior recognition method and device and camera | |
CN106501272A (en) | Machine vision scolding tin position detecting system | |
CN111709365A (en) | Automatic human motion posture detection method based on convolutional neural network | |
CN106651802A (en) | Machine vision tin soldering location detection method | |
CN110084777A (en) | A kind of micro parts positioning and tracing method based on deep learning | |
CN114187664B (en) | Rope skipping counting system based on artificial intelligence | |
Guo et al. | DeblurSLAM: A novel visual SLAM system robust in blurring scene | |
CN114863311A (en) | Automatic tracking method and system for inspection target of transformer substation robot | |
CN109815957A (en) | A kind of character recognition method based on color image under complex background | |
KR20110002338A (en) | Humman detection appartus using haar-like fearture and method thereof | |
CN110334703B (en) | Ship detection and identification method in day and night image |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190802 |