WO2018020954A1 - 機械学習用データベース作成システム - Google Patents
機械学習用データベース作成システム Download PDFInfo
- Publication number
- WO2018020954A1 WO2018020954A1 PCT/JP2017/024099 JP2017024099W WO2018020954A1 WO 2018020954 A1 WO2018020954 A1 WO 2018020954A1 JP 2017024099 W JP2017024099 W JP 2017024099W WO 2018020954 A1 WO2018020954 A1 WO 2018020954A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- information
- dimensional
- unit
- data
- dimensional shape
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/05—Geographic models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/56—Information retrieval; Database structures therefor; File system structures therefor of still image data having vectorial format
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5854—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using shape and object relationship
-
- 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
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T15/00—3D [Three Dimensional] image rendering
- G06T15/10—Geometric effects
- G06T15/20—Perspective computation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/64—Three-dimensional objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Definitions
- the present invention relates to a machine learning database creation system.
- mining work machines such as excavators and dump trucks are used for mining and transporting earth and sand.
- Mining work machines used in mines are required to be unmanned from the viewpoint of safety and cost reduction.
- dump trucks since the amount of earth and sand transported per unit time is directly linked to the progress of mining, efficient operation is required. Therefore, in order to efficiently transport a large amount of earth and sand outside the mining site, a mining system using an autonomously traveling dump truck capable of continuous operation is required.
- an obstacle detection device such as a millimeter wave radar, a laser sensor, a camera, or a stereo camera
- the millimeter wave radar has a high environmental resistance capable of operating even when dust or rain occurs, and has a high measurement distance performance.
- a stereo camera or a laser sensor can measure a three-dimensional shape, an obstacle on the road can be detected with high accuracy.
- machine learning may be used in recent years.
- machine learning a large amount of data from each sensor is collected to analyze trends and determine parameters.
- the threshold value was often manually designed based on design data and data obtained from verification experiments.
- this method is based on the experience of the designer, has low reliability, and requires a large number of design steps.
- parameter design is performed using machine learning.
- An example is a detection system for automobiles using a camera for a collision prevention system for passenger cars.
- other vehicles are photographed at various places and dates with cameras attached to the target vehicle, and the photographed image data is collected.
- teacher data indicating which part of the photographed image data is a vehicle to be detected by the system is created.
- the teacher data is often manually created for each piece of image data.
- the object recognition system using machine learning has some problems in implementation.
- One problem is the cost of collecting a large amount of image data.
- it is necessary to prepare a large amount of image data and teacher data for learning. If similar information is not given as learning data, it cannot be recognized.
- image data obtained by photographing the car from behind and teacher data are given, it is difficult to detect the vehicle when the front part of the car is visible in this system. For this reason, in order to develop an object recognition system that can detect automobiles of any posture, it is necessary to collect image data of automobiles of any posture when collecting image data for machine learning.
- teacher data is often manually created for each piece of image data.
- this is a method in which an area occupied by a car is designated by a rectangle or the like for a large amount of image data taken in advance, and the designated area is given as teacher data.
- an object recognition system based on machine learning requires tens of thousands to millions of pairs of image data and teacher data, and thus it takes a lot of cost to create teacher data for machine learning. Yes.
- Patent Document 1 a plurality of pieces of image information including an input image and a teacher image as an expected value of image processing for the input image are generated according to a scenario described in a program code, and the generated pieces of learning information are An information processing apparatus having a machine learning module that synthesizes an image processing algorithm by machine learning is disclosed.
- the present invention has been made in view of the above-described problems, and an object thereof is to provide a machine learning database creation system that can automatically create a large amount of virtual image data and teacher data.
- a 3D shape data input unit for inputting the 3D shape information of the terrain and building acquired by the 3D shape information measuring means, a 3D simulator unit for automatically recognizing and classifying the environment information from the 3D shape information,
- a machine learning database creation system comprising: a teacher data output unit that outputs virtual sensor data and teacher data based on environmental information recognized by a three-dimensional simulator unit and sensor parameters of a sensor.
- a machine learning database creation system capable of automatically creating a large amount of virtual image data and teacher data can be provided. Problems, configurations, and effects other than those described above will be clarified by the following description of embodiments.
- the present embodiment is an example in which a machine learning database is created using the present invention, and an object recognition system is configured by a machine learning system learned by the database.
- the machine learning database is used, for example, for object detection for an external sensing system for autonomously traveling vehicles.
- the machine learning database by inputting the three-dimensional shape data of the environment and the three-dimensional shape data of the target object to be detected, a scene is automatically created in the three-dimensional simulator unit, and virtual image data And teacher data can be automatically generated without human intervention. Accordingly, it is possible to provide a machine learning database automatic generation system capable of providing learning information such as image data and teacher data necessary for implementing an object recognition system using machine learning at a low cost. .
- the position and shape of the landmark for calibration are accurately measured and the vehicle position is estimated by, for example, a three-dimensional shape information acquisition unit using UAV or the like.
- the position between the sensors and the position between the sensor vehicles can be estimated and corrected with high accuracy. Thereby, it becomes possible to operate the obstacle detection system using the sensor soundly.
- FIG. 1 is a diagram showing the configuration of this embodiment.
- FIG. 2 is a diagram showing an example of a detailed configuration of an embodiment of the present invention.
- FIG. 3 is an example of a configuration for acquiring three-dimensional shape information of terrain.
- the three-dimensional shape data input unit 11 acquires the three-dimensional shape information of the terrain of the measurement target 3, and subsequently enters the vehicle information and sensor information into the three-dimensional simulator unit 12, thereby enabling the virtual space of the environment. Is generated.
- the teacher data output unit 13 provides virtual image data and a teacher data group to the machine learning system as a database of teacher data. Accordingly, it is possible to construct an object recognition system in which the machine learning system can learn from the virtual image data and the teacher data and can recognize the memorized measurement target object.
- FIG. 1 a detailed configuration of the present embodiment is shown in FIG.
- this embodiment will be described based on this configuration.
- the 3D shape data of the measurement object 3 is given to the topographic 3D shape data input unit 111 of the 3D shape data input unit 11 using the environment 3D information acquisition unit 45.
- a method for measuring a three-dimensional shape for example, a method of attaching a sensor such as an aerial camera 21 or a lidar to a UAV 2 (unmanned aircraft) as shown in FIG.
- the environment three-dimensional shape information acquisition means 45 includes, for example, an unmanned aircraft, a manned aircraft, an artificial satellite, a camera, a Lidar, a millimeter wave radar, an ultrasonic sensor, and similar environmental shapes or environmental luminance and color information.
- a sensor capable of acquiring temperature information is attached, measured, and acquired, but this configuration is not limited as long as three-dimensional shape information can be acquired.
- the three-dimensional shape data of the environment can be acquired even in a configuration in which a camera, a Lidar, and a GPS are attached to an automobile.
- the measurement object 3 when acquiring the three-dimensional shape information of the topography of the measurement target 3 using the UAV 2 and the aerial camera 21, in this configuration, first, the measurement object 3 is first caused to fly over. At this time, the measurement object 3 is continuously photographed by the aerial camera 21. At this time, it is desirable to shoot so that the captured image overlaps with the other captured images about 80% before and after, and about 60% lateral.
- FIG. 4 shows a method for acquiring three-dimensional shape information from an image photographed using UAV2.
- FIG. 4 is an example of a method for generating a three-dimensional shape of the terrain.
- the 3D point cloud generation unit 41 which is an algorithm implemented on the 3D reconstruction computer 4, using the image 211 captured by the UAV 2, uses Structure from Motion (SfM) and Multi View Stereo (MVS).
- SfM Structure from Motion
- MVS Multi View Stereo
- the three-dimensional shape information of the measurement object 3 can be obtained as point cloud information.
- the surface generation unit 42 meshes it to generate three-dimensional surface information having texture information and surface normal vector information.
- the three-dimensional shape information is stored in the three-dimensional shape information storage unit 43. Since these techniques are known techniques, they are omitted here. Thus, the three-dimensional terrain shape data 44 to be measured was obtained.
- the object three-dimensional information acquisition unit 46 may have a configuration similar to the environment three-dimensional information acquisition unit 45.
- means for acquiring three-dimensional shape information of an object by using a monocular camera or a plurality of cameras and SfM and MVS, or a method of measuring using Lidar can be considered. Since these techniques are also known, they are omitted here.
- this information is given as information having actual scale information such as a meter. Will also be given.
- the input object information is a vehicle
- the tire portion is set as the downward direction
- the roof direction is set as the upward direction of the object.
- a plurality of types of three-dimensional shape information provided from the object three-dimensional information acquisition means 46 may be used.
- the object three-dimensional shape data input unit 112 includes a three-dimensional shape of the dump truck, the excavator, and the worker. Enter the data.
- the terrain 3D shape data input unit 111 passes the received 3D shape information of the terrain to the 3D environment recognition unit 121. Based on this information, the three-dimensional environment recognition unit 121 automatically recognizes the received environment information of the terrain. As an example, a method in which the three-dimensional environment recognition unit 121 extracts a travelable area of an arbitrary vehicle as environment information from the three-dimensional shape information of the terrain will be described with reference to FIG. FIG. 5 is an example of a processing procedure of the three-dimensional environment recognition unit.
- the three-dimensional shape information of the A mine is input here (S11).
- the three-dimensional shape information is received as three-dimensional point group information.
- the three-dimensional shape information of the object to be detected is acquired (S12).
- information on the subject vehicle and sensor information are acquired (S13).
- information on the vehicle type A is given here as information on the own vehicle.
- the information to be given includes the shape of the host vehicle, the speed range when traveling, the climbing performance of uphill and downhill, and the climbing performance of steps and obstacles.
- the sensor information is the type of sensor mounted on the vehicle to recognize obstacles and the measurement performance. For example, when the camera is targeted as a sensor, the camera resolution, frame rate, lens focal length, distortion, etc. Sensor internal position information such as camera installation position and angle.
- the travelable area of the host vehicle is estimated from the obtained terrain information.
- a normal vector is calculated for each point based on each point of the three-dimensional point group and surrounding points (S14).
- the surrounding points with respect to an arbitrary point serving as a reference for calculating the normal vector are within a distance ⁇ 1 from the arbitrary point.
- the distance ⁇ 1 is a threshold value set in advance by the user.
- the calculated normal vector of each point is compared with the gravity vector.
- a point cloud having a normal vector tilted by ⁇ or more is deleted (S15).
- a clustering process is performed on the remaining point groups based on the Euclidean distance between the point groups (S16).
- points are clustered based on a preset threshold value ⁇ 2. For example, when it is determined that a point within the threshold ⁇ 2 is connected to an arbitrary point A and a point B, and a point that is more than the threshold ⁇ 2 is not connected, the distance between the arbitrary point A and the arbitrary point B is Even when the distance is greater than or equal to the threshold ⁇ 2, it is assumed that the point A and the point B are classified into the same class when they can reach an arbitrary point B by passing through a point within a distance within the other threshold ⁇ 2. .
- FIG. 6 is an example of a recognition result of the three-dimensional environment recognition unit.
- the automatic scenario creation unit 122 has a role of creating a scene from the obtained 3D shape information of the terrain and 3D shape information of the object. For example, it is assumed that dump truck object three-dimensional shape data 1121 as shown in FIG. 7 is given as a detection target of the object recognition system.
- FIG. 7 is an example of object three-dimensional shape data handled in an embodiment of the present invention.
- the automatic scenario creation unit 122 determines where the dump truck can exist from the dump truck, the three-dimensional shape data of the terrain given from the three-dimensional environment recognition unit 121, and the environment information. To do. For example, when one point is selected at random from the point group determined to be a travelable area by the three-dimensional environment recognition unit 121 and a dump truck is arranged there, the footprint of the dump truck deviates from the travelable area. Judge whether to do. If it deviates, the point is selected again at random, and if it does not deviate, it is determined to place a dump truck at that point.
- the scenario automatic creation unit 122 gives the information to the three-dimensional virtual space generation unit 125, and virtually arranges the dump truck based on the scenario set by the scenario automatic creation unit 122.
- An example of this is shown in FIG. FIG. 8 is an example of a scene generated by the three-dimensional virtual space generation unit.
- the object 3D shape data 1121 of the dump truck is synthesized with the 3D landform shape data 44. Further, the three-dimensional shape data 1121 of the dump truck is arranged on the travelable area. Based on the above processing, the three-dimensional simulator unit 12 generates a three-dimensional virtual space.
- the teacher data generation unit 131 of the teacher data output unit 13 Based on the information of the finally generated three-dimensional virtual space, the teacher data generation unit 131 of the teacher data output unit 13 generates teacher data, and the virtual sensor data generation unit 132 generates virtual sensor data.
- the position of the host vehicle in the three-dimensional virtual space is determined based on the vehicle information input in the vehicle parameter input unit 123 of the three-dimensional simulator unit 12. This is the same as the method of arranging the object three-dimensional shape data 1121 described in the automatic scenario creation unit 122, but the footprint of the object three-dimensional shape data 1121 arranged in advance and the footprint of the own vehicle are superimposed. If so, redo the point selection.
- virtual sensor data is generated in the virtual sensor data generation unit 132 according to the parameter input in the sensor parameter input unit 124 of 124.
- the input sensor data is for a camera
- a two-dimensional image that the camera can acquire is obtained by perspective projection conversion based on the installation position of the camera, the performance of the image element, and the performance and distortion of the lens.
- FIG. 9 is an example of virtual sensor data generated by the virtual sensor data generation unit.
- teacher data corresponding to the virtual sensor data 1321 is created.
- Environment information recognized by the three-dimensional environment recognition unit 121 is used to create the teacher data.
- the environment information is stored for each pixel of the 2D image obtained as the virtual sensor data.
- Teacher data is generated in the teacher data generation unit 131.
- FIG. 10 is an example of teacher data generated by the teacher data generation unit.
- FIG. 11 shows an example of a method for operating the object recognition algorithm using the generated virtual sensor data and teacher data.
- the virtual sensor data 1321 and the teacher data 1311 are given to the machine learning system 511 of the machine learning computer 5 to perform machine learning.
- a machine learning method used here Support Vector Machine, Boosting, a neural network, or an advanced method thereof can be considered. Since these methods are known techniques, they are omitted here.
- the obtained learning result 52 is given as a parameter of the object recognition algorithm 62.
- an appropriate feature amount for recognizing an object to be detected, a threshold necessary for object recognition using the feature amount, and the like can be considered.
- the object recognition algorithm 62 to which this parameter has been input detects a learned object or a similar object from the information obtained from the vehicle external sensor 61 and passes the information to the detection result output unit 63. An example is shown in FIG. FIG. 12 is an example of a detection result by the object recognition algorithm.
- the position of the vehicle ahead of the host vehicle is displayed as the detection result 71 on the display 7 installed in the vehicle.
- a method of notifying with an alarm when the subject vehicle is too close to the target object may be considered.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Geometry (AREA)
- Data Mining & Analysis (AREA)
- Remote Sensing (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Graphics (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Library & Information Science (AREA)
- Astronomy & Astrophysics (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Image Analysis (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Image Processing (AREA)
Abstract
Description
111 地形3次元形状データ入力部
112 物体3次元形状データ入力部
1121 物体3次元形状データ
12 3次元シミュレータ部
121 3次元環境認識部
122 シナリオ自動作成部
123 車両パラメータ入力部
124 センサパラメータ入力部
125 3次元仮想空間生成部
13 教師データ出力部
131 教師データ生成部
1311 教師データ
132 仮想センサデータ生成部
1321 仮想センサデータ
2 UAV
21 空撮用カメラ
211 撮影した画像
3 計測対象
4 3次元復元用計算機
41 3次元点群生成部
42 サーフェイス生成部
43 3次元形状情報格納部
44 3次元地形形状データ
45 環境3次元情報取得手段
46 物体3次元情報取得手段
5 機械学習用計算機
51 機械学習システム
52 学習結果
61 車両用外界センサ
62 物体認識アルゴリズム
63 検出結果出力部
7 ディスプレイ
71 検出結果
Claims (3)
- 3次元形状情報計測手段により取得した地形や建物の3次元形状情報を入力する3次元形状データ入力部と、
環境情報を前記3次元形状情報から自動的に認識し分類を行う3次元シミュレータ部と、
前記3次元シミュレータ部が認識した前記環境情報およびセンサのセンサパラメータに基づき、仮想センサデータおよび教師データを出力する教師データ出力部と、
を有する機械学習用データベース作成システム。 - 請求項1の機械学習用データベース作成システムにおいて、
前記3次元形状データ入力部は、
前記3次元形状情報を入力する地形3次元形状データ入力部と、
任意の物体の3次元形状データを入力する物体3次元形状データ入力部と、を有し、
前記3次元シミュレータ部は、前記地形3次元形状データ入力部と前記3次元形状データ入力部の情報を統合し、仮想空間を生成する3次元仮想空間生成部を有する
機械学習用データベース作成システム。 - 請求項1の機械学習用データベース作成システムにおいて、
前記3次元シミュレータ部は、前記地形3次元形状データ入力部で得られた前記3次元形状情報と、3次元環境認識部によって抽出された前記環境情報と前記物体3次元形状データ入力部が得た前記物体3次元形状情報と、に基づき、前記物体3次元形状情報と前記物体3次元形状情報の相対位置をランダムに作成するシナリオ自動作成部を有する機械学習用データベース作成システム。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2018529463A JP6770577B2 (ja) | 2016-07-29 | 2017-06-30 | 機械学習用データベース作成システム |
AU2017302833A AU2017302833B2 (en) | 2016-07-29 | 2017-06-30 | Database construction system for machine-learning |
US16/310,966 US10878288B2 (en) | 2016-07-29 | 2017-06-30 | Database construction system for machine-learning |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2016-149184 | 2016-07-29 | ||
JP2016149184 | 2016-07-29 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2018020954A1 true WO2018020954A1 (ja) | 2018-02-01 |
Family
ID=61016507
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2017/024099 WO2018020954A1 (ja) | 2016-07-29 | 2017-06-30 | 機械学習用データベース作成システム |
Country Status (4)
Country | Link |
---|---|
US (1) | US10878288B2 (ja) |
JP (1) | JP6770577B2 (ja) |
AU (1) | AU2017302833B2 (ja) |
WO (1) | WO2018020954A1 (ja) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2019145953A (ja) * | 2018-02-19 | 2019-08-29 | 国立大学法人 東京大学 | 作業車両の表示システム及び生成方法 |
JP2019172010A (ja) * | 2018-03-28 | 2019-10-10 | 株式会社豊田中央研究所 | 車両の走行を制御する制御装置、制御方法、及び、制御プログラム |
KR20190119691A (ko) * | 2018-04-12 | 2019-10-23 | 동국대학교 산학협력단 | 무차익 옵션 가격 예측 방법 |
WO2020066209A1 (ja) * | 2018-09-27 | 2020-04-02 | 株式会社堀場製作所 | 粒子解析用データ生成方法、粒子解析用データ生成プログラム、及び粒子解析用データ生成装置 |
WO2020067204A1 (ja) * | 2018-09-27 | 2020-04-02 | 株式会社Preferred Networks | 学習用データ作成方法、機械学習モデルの生成方法、学習用データ作成装置及びプログラム |
JP2020060907A (ja) * | 2018-10-09 | 2020-04-16 | 株式会社パスコ | 避雷保護範囲生成システムおよびプログラム |
US11132826B2 (en) | 2019-05-16 | 2021-09-28 | Caterpillar Inc. | Artificial image generation for training an object detection system |
WO2022080356A1 (ja) * | 2020-10-13 | 2022-04-21 | 株式会社Preferred Networks | 訓練データ生成装置、訓練済みモデル生成装置、物体認識支援システム、物体認識システム、訓練データ生成方法、訓練済みモデル生成方法、物体認識支援方法及び物体認識方法 |
US20220165058A1 (en) * | 2019-04-09 | 2022-05-26 | Komatsu Ltd. | Information processing device, information processing method, trained model generation method, system, and learning data set |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP6845106B2 (ja) * | 2017-07-21 | 2021-03-17 | 株式会社タダノ | 点群データのクラスタリング方法、ガイド情報表示装置およびクレーン |
US10768304B2 (en) * | 2017-12-13 | 2020-09-08 | Luminar Technologies, Inc. | Processing point clouds of vehicle sensors having variable scan line distributions using interpolation functions |
JP6900897B2 (ja) * | 2017-12-25 | 2021-07-07 | コベルコ建機株式会社 | 建設機械の障害物検出装置 |
US11046519B2 (en) * | 2019-02-25 | 2021-06-29 | Rehrig Pacific Company | Delivery system |
JP7384898B2 (ja) * | 2019-03-13 | 2023-11-21 | 日本電気株式会社 | 情報処理装置、駆動制御方法およびコンピュータプログラム |
CN112927345B (zh) * | 2021-01-20 | 2022-11-15 | 中国地震应急搜救中心 | 一种地震救援现场环境的监控方法和设备 |
CN115422386A (zh) * | 2022-09-29 | 2022-12-02 | 南方电网调峰调频发电有限公司西部检修试验分公司 | 一种机器学习数据分析框架信息库搭建方法及装置 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2000163685A (ja) * | 1998-11-30 | 2000-06-16 | Fuji Electric Co Ltd | 交通流監視システム |
JP2006172099A (ja) * | 2004-12-15 | 2006-06-29 | Mitsubishi Electric Corp | 変化領域認識装置および変化認識システム |
WO2011086889A1 (ja) * | 2010-01-12 | 2011-07-21 | 日本電気株式会社 | 特徴点選択システム、特徴点選択方法および特徴点選択プログラム |
JP2014157509A (ja) * | 2013-02-15 | 2014-08-28 | Canon Inc | 情報処理装置、情報処理方法 |
JP2016006616A (ja) * | 2014-06-20 | 2016-01-14 | ヤフー株式会社 | 学習装置、学習方法及び学習プログラム |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103953209B (zh) * | 2012-12-12 | 2017-06-09 | 保利集团澳门有限公司 | 水池系统及建造使用方法 |
JP2014147509A (ja) | 2013-01-31 | 2014-08-21 | Soka Univ | 被洗浄物の洗浄方法及び洗浄装置 |
JP2016071597A (ja) | 2014-09-30 | 2016-05-09 | ソニー株式会社 | 情報処理装置、情報処理方法及びプログラム |
US9970756B2 (en) * | 2015-10-06 | 2018-05-15 | Bridger Photonics, Inc. | High-sensitivity gas-mapping 3D imager and method of operation |
-
2017
- 2017-06-30 AU AU2017302833A patent/AU2017302833B2/en active Active
- 2017-06-30 US US16/310,966 patent/US10878288B2/en active Active
- 2017-06-30 JP JP2018529463A patent/JP6770577B2/ja active Active
- 2017-06-30 WO PCT/JP2017/024099 patent/WO2018020954A1/ja active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2000163685A (ja) * | 1998-11-30 | 2000-06-16 | Fuji Electric Co Ltd | 交通流監視システム |
JP2006172099A (ja) * | 2004-12-15 | 2006-06-29 | Mitsubishi Electric Corp | 変化領域認識装置および変化認識システム |
WO2011086889A1 (ja) * | 2010-01-12 | 2011-07-21 | 日本電気株式会社 | 特徴点選択システム、特徴点選択方法および特徴点選択プログラム |
JP2014157509A (ja) * | 2013-02-15 | 2014-08-28 | Canon Inc | 情報処理装置、情報処理方法 |
JP2016006616A (ja) * | 2014-06-20 | 2016-01-14 | ヤフー株式会社 | 学習装置、学習方法及び学習プログラム |
Non-Patent Citations (1)
Title |
---|
NAOYA KOKETSU ET AL.: "Improvement in Human Detection by Generation Addition Training of Human Silhouette", IPSJ SIG NOTES 2010, 15 October 2010 (2010-10-15) * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2019145953A (ja) * | 2018-02-19 | 2019-08-29 | 国立大学法人 東京大学 | 作業車両の表示システム及び生成方法 |
JP7232437B2 (ja) | 2018-02-19 | 2023-03-03 | 国立大学法人 東京大学 | 作業車両の表示システム及び生成方法 |
JP2019172010A (ja) * | 2018-03-28 | 2019-10-10 | 株式会社豊田中央研究所 | 車両の走行を制御する制御装置、制御方法、及び、制御プログラム |
JP7000952B2 (ja) | 2018-03-28 | 2022-01-19 | 株式会社豊田中央研究所 | 車両の走行を制御する制御装置、制御方法、及び、制御プログラム |
KR102120655B1 (ko) * | 2018-04-12 | 2020-06-18 | 동국대학교 산학협력단 | 무차익 옵션 가격 예측 방법 |
KR20190119691A (ko) * | 2018-04-12 | 2019-10-23 | 동국대학교 산학협력단 | 무차익 옵션 가격 예측 방법 |
WO2020066209A1 (ja) * | 2018-09-27 | 2020-04-02 | 株式会社堀場製作所 | 粒子解析用データ生成方法、粒子解析用データ生成プログラム、及び粒子解析用データ生成装置 |
JPWO2020066209A1 (ja) * | 2018-09-27 | 2021-09-09 | 株式会社堀場製作所 | 粒子解析用データ生成方法、粒子解析用データ生成プログラム、及び粒子解析用データ生成装置 |
US12105009B2 (en) | 2018-09-27 | 2024-10-01 | Horiba, Ltd. | Method for generating data for particle analysis, program for generating data for particle analysis, and device for generating data for particle analysis |
JP7538718B2 (ja) | 2018-09-27 | 2024-08-22 | 株式会社堀場製作所 | 粒子解析用データ生成方法、粒子解析用データ生成プログラム、及び粒子解析用データ生成装置 |
WO2020067204A1 (ja) * | 2018-09-27 | 2020-04-02 | 株式会社Preferred Networks | 学習用データ作成方法、機械学習モデルの生成方法、学習用データ作成装置及びプログラム |
JP2020060907A (ja) * | 2018-10-09 | 2020-04-16 | 株式会社パスコ | 避雷保護範囲生成システムおよびプログラム |
JP7184249B2 (ja) | 2018-10-09 | 2022-12-06 | 株式会社パスコ | 避雷保護範囲生成システムおよびプログラム |
US20220165058A1 (en) * | 2019-04-09 | 2022-05-26 | Komatsu Ltd. | Information processing device, information processing method, trained model generation method, system, and learning data set |
CN113826112A (zh) * | 2019-05-16 | 2021-12-21 | 卡特彼勒公司 | 用于训练对象检测系统的人工图像生成 |
US11132826B2 (en) | 2019-05-16 | 2021-09-28 | Caterpillar Inc. | Artificial image generation for training an object detection system |
WO2022080356A1 (ja) * | 2020-10-13 | 2022-04-21 | 株式会社Preferred Networks | 訓練データ生成装置、訓練済みモデル生成装置、物体認識支援システム、物体認識システム、訓練データ生成方法、訓練済みモデル生成方法、物体認識支援方法及び物体認識方法 |
Also Published As
Publication number | Publication date |
---|---|
US20200202175A1 (en) | 2020-06-25 |
JP6770577B2 (ja) | 2020-10-14 |
AU2017302833A1 (en) | 2019-01-17 |
US10878288B2 (en) | 2020-12-29 |
JPWO2018020954A1 (ja) | 2019-04-04 |
AU2017302833A8 (en) | 2019-03-14 |
AU2017302833B2 (en) | 2020-11-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2018020954A1 (ja) | 機械学習用データベース作成システム | |
US11691648B2 (en) | Drivable surface identification techniques | |
CN110531753B (zh) | 自主车辆的控制系统、控制方法和控制器 | |
KR102572219B1 (ko) | 자율 주행을 위한 항법 정보의 융합 프레임워크(fusion framework) 및 배치 정렬(batch alignment) | |
Jebamikyous et al. | Autonomous vehicles perception (avp) using deep learning: Modeling, assessment, and challenges | |
JP6663406B2 (ja) | 車両制御装置、車両制御方法、およびプログラム | |
CN111918053A (zh) | 车辆图像验证 | |
Shim et al. | An autonomous driving system for unknown environments using a unified map | |
CN111442776A (zh) | 顺序地面场景图像投影合成与复杂场景重构的方法及设备 | |
WO2020185489A1 (en) | Sensor validation using semantic segmentation information | |
EP4187523A1 (en) | Systems and methods for curb detection and pedestrian hazard assessment | |
DE112019001657T5 (de) | Signalverarbeitungsvorrichtung und signalverarbeitungsverfahren, programm und mobiler körper | |
KR20190080885A (ko) | 차로 병합 및 차로 분리의 항법을 위한 시스템 및 방법 | |
US20200082182A1 (en) | Training data generating method for image processing, image processing method, and devices thereof | |
KR20200127219A (ko) | 검출된 배리어에 기반한 차량의 항법 | |
KR20180055292A (ko) | 다중 라이다 좌표계 통합 방법 | |
EP3842751A1 (en) | System and method of generating high-definition map based on camera | |
CN114419098A (zh) | 基于视觉变换的运动目标轨迹预测方法及装置 | |
EP3555854B1 (en) | A method of tracking objects in a scene | |
KR102618680B1 (ko) | 영상과 라이다를 이용한 실시간 3차원 물체 인식 및 추적하는 시스템 | |
López et al. | Computer vision in vehicle technology: Land, sea, and air | |
US20230230317A1 (en) | Method for generating at least one ground truth from a bird's eye view | |
Jung et al. | Intelligent hybrid fusion algorithm with vision patterns for generation of precise digital road maps in self-driving vehicles | |
Bhatlawande et al. | LIDAR based Detection of Small Vehicles | |
CN117836818A (zh) | 信息处理装置、信息处理系统、模型以及模型的生成方法 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 17833963 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2018529463 Country of ref document: JP Kind code of ref document: A |
|
ENP | Entry into the national phase |
Ref document number: 2017302833 Country of ref document: AU Date of ref document: 20170630 Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 17833963 Country of ref document: EP Kind code of ref document: A1 |