CN109871850A - A kind of classification method of the mobile lidar data based on neural network model - Google Patents
A kind of classification method of the mobile lidar data based on neural network model Download PDFInfo
- Publication number
- CN109871850A CN109871850A CN201910053794.0A CN201910053794A CN109871850A CN 109871850 A CN109871850 A CN 109871850A CN 201910053794 A CN201910053794 A CN 201910053794A CN 109871850 A CN109871850 A CN 109871850A
- Authority
- CN
- China
- Prior art keywords
- neural network
- training
- data
- sample
- unsupervised
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
- G01S17/88—Lidar systems specially adapted for specific applications
- G01S17/93—Lidar systems specially adapted for specific applications for anti-collision purposes
- G01S17/931—Lidar systems specially adapted for specific applications for anti-collision purposes of land vehicles
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
- G01S17/88—Lidar systems specially adapted for specific applications
- G01S17/89—Lidar systems specially adapted for specific applications for mapping or imaging
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Electromagnetism (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Traffic Control Systems (AREA)
Abstract
The classification method for the mobile lidar data based on neural network model that the embodiment of the invention provides a kind of, comprising: generate sample set using three-dimensional laser radar point cloud data;Unsupervised mark is carried out to the sample in sample set based on priori knowledge, parameter pre-training is carried out to neural network model using unsupervised labeled data;Arameter optimization is carried out to the neural network model after parameter pre-training using the data marked by hand, obtains the neural network classification model of the mobile lidar data of training completion;The neural network classification model for the mobile lidar data completed using the training classifies to mobile lidar data to be processed.The embodiment of the present invention carries out pre-training to neural network by the priori knowledge provided using the mankind, to reduce dependence of the neural network in the training process to manual labeled data, improves network training efficiency.
Description
Technical field
The present invention relates to the semantic classification technique fields of laser radar data, more particularly to one kind to be based on neural network model
Mobile lidar data classification method.
Background technique
With the fast development of automatic Pilot technology, three-dimensional laser radar is answered extensively as a kind of main awareness apparatus
For having important application demand in fields such as communications and transportation, national defense safeties in automated driving system.It is sensed compared to camera
Device, laser radar directly export distance measure, and can all weather operations, these characteristics make laser radar data big
Range applications are in the perception part of automatic Pilot platform.With the reduction of sensor cost, the semanteme based on laser radar data
Classification receives more and more attention.For the semantic classification of laser data, conventional method relies on the priori knowledge of the mankind, by hand
Design feature and classifier carry out labeling, the advantage of such methods to the data cell of input in the way of machine learning
It is that the interpretation of model, disadvantage are that manual feature is poor to the adaptability of environment.
Neural network method brings new think of to the semantic classification of laser radar data in the success of field of image recognition
Road, its advantage is that feature extraction is automatically carried out in a manner of data-driven, to remove in conventional method to manual feature
Dependence;The disadvantage is that have a large amount of demand to manual labeled data, it is fine to three-dimensional laser radar data compared to image
It is relatively difficult for changing mark.For laser radar data, existing neural network method focus on network structure design and
The production of fine craft labeled data, and seldom consider the importance of mankind's priori knowledge.
Summary of the invention
The classification method for the mobile lidar data based on neural network model that the embodiment provides a kind of,
To overcome the deficiencies of existing technologies.
To achieve the goals above, this invention takes following technical solutions.
A kind of classification method of the mobile lidar data based on neural network model, comprising:
Sample set is generated using three-dimensional laser radar point cloud data;
Unsupervised mark is carried out to the sample in sample set based on priori knowledge, using unsupervised labeled data to nerve
Network model carries out parameter pre-training;
Arameter optimization is carried out to the neural network model after parameter pre-training using the data marked by hand, obtains having trained
At mobile lidar data neural network classification model;
The neural network classification model for the mobile lidar data completed using the training vehicle-mounted is swashed to be processed
Optical radar data are classified.
Preferably, the utilization three-dimensional laser radar point cloud data generates sample set, comprising:
Known three-dimensional laser radar point cloud data is converted into two-dimensional distance image;
RegionGrow is carried out to the two-dimensional distance image to divide to obtain segmentation block;
The three-dimensional point cloud for facing domain around taking out centered on each segmentation block generates each sample;
Gather each sample and generates sample set.
Preferably, described to carry out unsupervised mark to the sample in sample set based on priori knowledge, utilization is unsupervised
Labeled data carries out parameter pre-training to neural network model, comprising:
For the characteristic of laser radar data, unsupervised segmentation is carried out to sample in sample set using priori knowledge, often
The corresponding segmentation block of a sample, i.e., be equivalent to sample classification the classification of segmentation block, unsupervised segmentation is by utilizing priori
Knowledge classifies to the height of segmentation block and width characteristics identification, generates corresponding mark for sample xObtain unsupervised mark
Infuse dataUtilize the data of unsupervised markNeural network parameter pre-training is carried out, in the training of neural network
Optimize loss function in the processThe function is defined as cross entropy, is iterated by the way of stochastic gradient descent
Training, network initial parameter are obtained by random initializtion, and the value of loss function can gradually decrease in the training process of neural network,
Deconditioning after penalty values are stablized, obtains optimal network parameter
Preferably, described that parameter tune is carried out to the neural network model after parameter pre-training using the data marked by hand
It is excellent, comprising:
Parameter migration is carried out first, uses the parameter model of pre-trainingAnother network is initialized, then into
Row neural network model parameter tuning, majorized function becomes in the evolutionary processThe function is defined as intersecting
Entropy is iterated training by the way of stochastic gradient descent, but the initial position of gradient decline isLetter should be lost in the process
Several values can gradually decrease, deconditioning after penalty values are stablized, completion tuning acquisition θ, after neural network model parameter tuning
Obtain the neural network classification model of the mobile lidar data of training completion.
As can be seen from the technical scheme provided by the above-mentioned embodiment of the present invention, the embodiment of the present invention using the mankind by being mentioned
The priori knowledge of confession carries out pre-training to neural network, to reduce neural network in the training process to manual labeled data
It relies on, improves network training efficiency.
The additional aspect of the present invention and advantage will be set forth in part in the description, these will become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, required use in being described below to embodiment
Attached drawing be briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this
For the those of ordinary skill of field, without creative efforts, it can also be obtained according to these attached drawings others
Attached drawing.
Fig. 1 is the classification side of the embodiment of the invention provides a kind of mobile lidar data based on neural network model
The flow diagram of method;
Fig. 2 is the flow diagram that the sample based on segmentation block generates;
Fig. 3 is Knowledge based engineering pre-training data product process schematic diagram;
Fig. 4 is Knowledge based engineering network training flow diagram.
Specific embodiment
Embodiments of the present invention are described below in detail, the example of the embodiment is shown in the accompanying drawings, wherein from beginning
Same or similar element or element with the same or similar functions are indicated to same or similar label eventually.Below by ginseng
The embodiment for examining attached drawing description is exemplary, and for explaining only the invention, and is not construed as limiting the claims.
Those skilled in the art of the present technique are appreciated that unless expressly stated, singular " one " used herein, " one
It is a ", " described " and "the" may also comprise plural form.It is to be further understood that being arranged used in specification of the invention
Diction " comprising " refer to that there are the feature, integer, step, operation, element and/or component, but it is not excluded that in the presence of or addition
Other one or more features, integer, step, operation, element, component and/or their group.It should be understood that when we claim member
Part is " connected " or when " coupled " to another element, it can be directly connected or coupled to other elements, or there may also be
Intermediary element.In addition, " connection " used herein or " coupling " may include being wirelessly connected or coupling.Wording used herein
"and/or" includes one or more associated any cells for listing item and all combinations.
Those skilled in the art of the present technique are appreciated that unless otherwise defined, all terms used herein (including technology art
Language and scientific term) there is meaning identical with the general understanding of those of ordinary skill in fields of the present invention.Should also
Understand, those terms such as defined in the general dictionary, which should be understood that, to be had and the meaning in the context of the prior art
The consistent meaning of justice, and unless defined as here, it will not be explained in an idealized or overly formal meaning.
In order to facilitate understanding of embodiments of the present invention, it is done by taking several specific embodiments as an example below in conjunction with attached drawing further
Explanation, and each embodiment does not constitute the restriction to the embodiment of the present invention.
The classification method for the mobile lidar data based on neural network model that the embodiment of the invention provides a kind of
Process flow, as shown in Figure 1, comprising the following steps:
Sample set is generated using three-dimensional laser radar point cloud data, as shown in Figure 2, comprising:
Known three-dimensional laser radar point cloud data is converted into two-dimensional distance image, two-dimensional distance image is carried out
RegionGrow (region growing) segmentation obtains segmentation block, and the three-dimensional point cloud for facing domain around taking-up centered on each segmentation block is raw
At each sample, gathers each sample and generate sample set.
Unsupervised mark is carried out to the sample in sample set based on priori knowledge, using unsupervised labeled data to nerve
Network model carries out parameter pre-training, comprising:
Unsupervised mark: for the characteristic of laser radar data, the priori knowledge of the mankind can be directly used for unsupervised point
Class.Such as: trunk (Trunk) is high and long, and people (People) is generally below 2 meters, and the width of vehicle (Car) is greater than the width of people,
The metope of building (Building) is substantially flat.For the integrality of sample label, unknown (Unknown) label is added.
The input data of neural network is made, it is directly higher to original point cloud classification complexity, under range image, believed using set
Breath is first split block generation, then will be greatly reduced the complexity of problem to segmentation block sort, therefore, a sample is defined as
The next segmentation block of range image and surrounding point cloud data.The corresponding segmentation block of each sample, as shown in Fig. 2, therefore
The classification of segmentation block is equivalent to sample classification.Unsupervised segmentation is using the height z and width w feature of segmentation block to segmentation block
Classify, as shown in figure 3, if the height of segmentation block is less than 2m (z < 2.0) and width in 0.2~1.5m (0.2 < w <
1.5) between, then being judged as the segmentation block is people (people), i.e. the corresponding sample label of the segmentation block is behaved;People herein
Height and width information are the priori knowledge of people.By the process, corresponding mark is generated for sample xSample and mark are equal
It can automatically complete, therefore can get a large amount of unsupervised labeled data
Neural network model parameter pre-training: the data of a large amount of unsupervised marks are utilizedCarry out parameter pre-training.
The training process of neural network needs to optimize loss functionThe function is defined as cross entropy, using under stochastic gradient
The mode of drop is iterated training (note: cross entropy and stochastic gradient descent are the most common modes in neural metwork training), net
Network initial parameter is obtained by random initializtion, and the value of loss function can gradually decrease during being somebody's turn to do, and is stopped after penalty values are stablized
Training, obtains optimal network parameterUnsupervised labeled data contains the priori knowledge of the mankind, carries out first to network
Then random initializtion is trained using unsupervised labeled data, as shown in Fig. 4 (a).
Arameter optimization is carried out to the neural network model after parameter pre-training using a small amount of data marked by hand, is instructed
Practice the neural network classification model for the mobile lidar data completed.
Neural network model parameter tuning: the tuning (mark by hand of parameter is carried out using a small amount of manual labeled data (x, y)
It is the operation to be labelled by people to initial data, which is typically all offline complete).As shown in figure 4, being joined first
Number migration: the parameter model of pre-training is usedAnother network is initialized, which can carry out parameter certain
It cuts, such as only to the parameter application of convolutional layerThen carry out network parameter tuning: majorized function becomes during being somebody's turn to doThe function is defined as cross entropy, training is iterated by the way of stochastic gradient descent, but gradient declines
Initial position beValue of loss function can gradually decrease during this, and deconditioning after penalty values are stablized completes tuning
θ is obtained, as shown in Fig. 4 (b).The nerve of the mobile lidar data of training completion is obtained after neural network model parameter tuning
Network class model.
The neural network classification model for the mobile lidar data completed using training is to vehicle-mounted laser thunder to be processed
Classify up to data.
In conclusion the embodiment of the present invention carries out pre-training to neural network by the priori knowledge provided using the mankind,
To reduce dependence of the neural network in the training process to manual labeled data, network training efficiency is improved.
Those of ordinary skill in the art will appreciate that: attached drawing is the schematic diagram of one embodiment, module in attached drawing or
Process is not necessarily implemented necessary to the present invention.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
In the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by anyone skilled in the art,
It should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with scope of protection of the claims
Subject to.
Claims (4)
1. a kind of classification method of the mobile lidar data based on neural network model characterized by comprising
Sample set is generated using three-dimensional laser radar point cloud data;
Unsupervised mark is carried out to the sample in sample set based on priori knowledge, using unsupervised labeled data to neural network
Model carries out parameter pre-training;
Arameter optimization is carried out to the neural network model after parameter pre-training using the data marked by hand, obtains training completion
The neural network classification model of mobile lidar data;
The neural network classification model for the mobile lidar data completed using the training is to vehicle-mounted laser thunder to be processed
Classify up to data.
2. classification method according to claim 1, which is characterized in that described is raw using three-dimensional laser radar point cloud data
At sample set, comprising:
Known three-dimensional laser radar point cloud data is converted into two-dimensional distance image;
RegionGrow is carried out to the two-dimensional distance image to divide to obtain segmentation block;
The three-dimensional point cloud for facing domain around taking out centered on each segmentation block generates each sample;
Gather each sample and generates sample set.
3. classification method according to claim 2, which is characterized in that it is described based on priori knowledge in sample set
Sample carries out unsupervised mark, carries out parameter pre-training to neural network model using unsupervised labeled data, comprising:
For the characteristic of laser radar data, unsupervised segmentation, each sample are carried out to sample in sample set using priori knowledge
One segmentation block of this correspondence is equivalent to sample classification the classification of segmentation block, unsupervised segmentation is by utilizing priori knowledge
Height and width characteristics identification to segmentation block are classified, and generate corresponding mark for sample xObtain unsupervised mark number
According toUtilize the data of unsupervised markNeural network parameter pre-training is carried out, in the training process of neural network
Middle optimization loss functionThe function is defined as cross entropy, and instruction is iterated by the way of stochastic gradient descent
To practice, network initial parameter is obtained by random initializtion, and the value of loss function can gradually decrease in the training process of neural network, when
Deconditioning after penalty values are stablized, obtains optimal network parameter
4. classification method according to claim 3, which is characterized in that the data that the utilization marks by hand are pre- to parameter
Neural network model after training carries out arameter optimization, comprising:
Parameter migration is carried out first, uses the parameter model of pre-trainingAnother network is initialized, nerve is then carried out
Network model arameter optimization, majorized function becomes in the evolutionary processThe function is defined as cross entropy, adopts
It is iterated training with the mode of stochastic gradient descent, but the initial position of gradient decline isLoss function during this
Value can gradually decrease, deconditioning after penalty values are stablized, and complete tuning and obtain θ, obtain after neural network model parameter tuning
The neural network classification model for the mobile lidar data that training is completed.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910053794.0A CN109871850A (en) | 2019-01-21 | 2019-01-21 | A kind of classification method of the mobile lidar data based on neural network model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910053794.0A CN109871850A (en) | 2019-01-21 | 2019-01-21 | A kind of classification method of the mobile lidar data based on neural network model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109871850A true CN109871850A (en) | 2019-06-11 |
Family
ID=66917840
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910053794.0A Withdrawn CN109871850A (en) | 2019-01-21 | 2019-01-21 | A kind of classification method of the mobile lidar data based on neural network model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109871850A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111537980A (en) * | 2020-07-08 | 2020-08-14 | 深圳市速腾聚创科技有限公司 | Laser radar parameter adjusting method and device and laser radar |
CN113870359A (en) * | 2021-09-23 | 2021-12-31 | 厦门大学 | Laser radar external parameter calibration method and device based on random gradient descent |
EP4170378A1 (en) * | 2021-10-20 | 2023-04-26 | Aptiv Technologies Limited | Methods and systems for processing radar sensor data |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104965819A (en) * | 2015-07-12 | 2015-10-07 | 大连理工大学 | Biomedical event trigger word identification method based on syntactic word vector |
CN104966097A (en) * | 2015-06-12 | 2015-10-07 | 成都数联铭品科技有限公司 | Complex character recognition method based on deep learning |
CN107124761A (en) * | 2017-03-10 | 2017-09-01 | 华南理工大学 | Merge PSO and SS ELM Wireless Location in Cellular Network method |
CN109145677A (en) * | 2017-06-15 | 2019-01-04 | 百度在线网络技术(北京)有限公司 | Obstacle detection method, device, equipment and storage medium |
-
2019
- 2019-01-21 CN CN201910053794.0A patent/CN109871850A/en not_active Withdrawn
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104966097A (en) * | 2015-06-12 | 2015-10-07 | 成都数联铭品科技有限公司 | Complex character recognition method based on deep learning |
CN104965819A (en) * | 2015-07-12 | 2015-10-07 | 大连理工大学 | Biomedical event trigger word identification method based on syntactic word vector |
CN107124761A (en) * | 2017-03-10 | 2017-09-01 | 华南理工大学 | Merge PSO and SS ELM Wireless Location in Cellular Network method |
CN109145677A (en) * | 2017-06-15 | 2019-01-04 | 百度在线网络技术(北京)有限公司 | Obstacle detection method, device, equipment and storage medium |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111537980A (en) * | 2020-07-08 | 2020-08-14 | 深圳市速腾聚创科技有限公司 | Laser radar parameter adjusting method and device and laser radar |
CN111999720A (en) * | 2020-07-08 | 2020-11-27 | 深圳市速腾聚创科技有限公司 | Laser radar parameter adjustment method, laser radar system, and computer storage medium |
US11768293B2 (en) | 2020-07-08 | 2023-09-26 | Suteng Innovation Technology Co., Ltd. | Method and device for adjusting parameters of LiDAR, and LiDAR |
US12055635B2 (en) | 2020-07-08 | 2024-08-06 | Suteng Innovation Technology Co., Ltd. | Method and device for adjusting parameters of LiDAR, and LiDAR |
CN113870359A (en) * | 2021-09-23 | 2021-12-31 | 厦门大学 | Laser radar external parameter calibration method and device based on random gradient descent |
CN113870359B (en) * | 2021-09-23 | 2024-07-19 | 厦门大学 | Laser radar external parameter calibration method and device based on random gradient descent |
EP4170378A1 (en) * | 2021-10-20 | 2023-04-26 | Aptiv Technologies Limited | Methods and systems for processing radar sensor data |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109902806B (en) | Method for determining target bounding box of noise image based on convolutional neural network | |
CN111274970B (en) | Traffic sign detection method based on improved YOLO v3 algorithm | |
CN109063723A (en) | The Weakly supervised image, semantic dividing method of object common trait is excavated based on iteration | |
CN109271926A (en) | Intelligent Radiation source discrimination based on GRU depth convolutional network | |
CN108288035A (en) | The human motion recognition method of multichannel image Fusion Features based on deep learning | |
CN109146944B (en) | Visual depth estimation method based on depth separable convolutional neural network | |
CN109871850A (en) | A kind of classification method of the mobile lidar data based on neural network model | |
CN110175615B (en) | Model training method, domain-adaptive visual position identification method and device | |
CN104850845A (en) | Traffic sign recognition method based on asymmetric convolution neural network | |
CN101964063B (en) | Method for constructing improved AdaBoost classifier | |
CN108647568B (en) | Grassland degradation automatic extraction method based on full convolution neural network | |
CN111126332B (en) | Frequency hopping signal classification method based on contour features | |
CN108009548A (en) | A kind of Intelligent road sign recognition methods and system | |
CN110276247A (en) | A kind of driving detection method based on YOLOv3-Tiny | |
CN112364721A (en) | Road surface foreign matter detection method | |
CN110807485B (en) | Method for fusing two-classification semantic segmentation maps into multi-classification semantic map based on high-resolution remote sensing image | |
CN104463194A (en) | Driver-vehicle classification method and device | |
CN102915453A (en) | Real-time feedback and update vehicle detection method | |
CN112395953A (en) | Road surface foreign matter detection system | |
CN108961270B (en) | Bridge crack image segmentation model based on semantic segmentation | |
Yu et al. | Obstacle detection with deep convolutional neural network | |
CN104850862A (en) | Classification method based on unit cost gain sensitive decision tree | |
CN110674845B (en) | Dish identification method combining multi-receptive-field attention and characteristic recalibration | |
CN106022382A (en) | Mixed-norm multiple indefinite kernel classification method for complex data | |
Suvetha et al. | Automatic Traffic Sign Detection System With Voice Assistant |
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 | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190611 |
|
WW01 | Invention patent application withdrawn after publication |