How to Perform Manual Classification for Deep Learning Using CloudCompare

Slide 1

Slide 1 text

How to Perform Manual Classification for Deep Learning Using CloudCompare ImVisionLabs Inc., CEO Kenta Itakura, Ph.D

Slide 2

Slide 2 text

Trend of new functions in CloudCompare 2 iPhone12 LiDAR  Interview of Dr. Daniel by Eugene in YouTube  One trend of new functions in CloudCompare is Manual Segmentation Interview with Daniel Girardeau-Montaut, Click 3D Episode 50 https://www.youtube.com/watch?v=sBpi1yHC4xA

Slide 3

Slide 3 text

Manual Classification tool in CloudCompare 3 iPhone12 LiDAR  Introduction to the cloud layers function for manual classification

Slide 4

Slide 4 text

Need for manual labeling 4 iPhone12 LiDAR  The figure below shows the automatic classification using deep learning Point cloud Automatic classification result  Annotation data is required for training classification models  Manual classification is needed for training data

Slide 5

Slide 5 text

Example of point cloud processing: Tree segmentation 5

Slide 6

Slide 6 text

上空から取得した樹木の解析について 6  より広範囲のデータの解析が可能  UAVレーザー測量により森林を計測し、樹木解析を実行 ※ Data from the Tokyo Metropolitan Government's Digital Twin Realization Project is used

Slide 7

Slide 7 text

Need for manual labeling 7 iPhone12 LiDAR  Manual classification is also required for evaluating segmentation results • Labeling is also needed to verify the accuracy of studies that count the number of trees Itakura, K., et al. (2021). Estimating tree structural parameters via automatic tree segmentation from LiDAR point cloud data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 15, 555-564.

Slide 8

Slide 8 text

Point cloud segmentation: Our work with Tepco 8  Use case of 3D point cloud data to improve the efficiency of patrols of power distribution facilities  Classification into Vegetation, Wire and Ground ◼ Ground ◼ Vegetation ◼ Wire Point cloud colored by elevation Auto Classification result

Slide 9

Slide 9 text

Point cloud segmentation: Our work with Mr. Takata 9  Automatically detect “traces of a path” from point clouds  The “traces of a path” leads to places where stones were once extracted for historical structures 高さごとに色分けした図自動分類した結果 Detected potential traces of a path Input point cloud

Slide 10

Slide 10 text

How to use CloudLayers: 10 iPhone12 LiDAR  Plugins -> Cloud layers iPhone12 LiDAR

Slide 11

Slide 11 text

How to use CloudLayers: 11 iPhone12 LiDAR  The panel is opened

Slide 12

Slide 12 text

How to use CloudLayers: 12 iPhone12 LiDAR  Select your target class

Slide 13

Slide 13 text

How to use CloudLayers: 13 iPhone12 LiDAR  button : Start Manual classification

Slide 14

Slide 14 text

How to use CloudLayers: 14 iPhone12 LiDAR  Circle radius can be changed  Use a smaller radius for painting small areas and a larger radius for larger areas

Slide 15

Slide 15 text

How to use CloudLayers: 15 iPhone12 LiDAR  Ground extraction made manual classification easy

Slide 16

Slide 16 text

How to use CloudLayers: 16 iPhone12 LiDAR  Ground extraction made manual classification easy

Slide 17

Slide 17 text

How to use CloudLayers: 17 iPhone12 LiDAR  Ground extraction made manual classification easy

Slide 18

Slide 18 text

Example of manual classification 18 iPhone12 LiDAR  Open data by Tokyo was manually labeled  Cloud layers function enabled efficient manual classification

Slide 19

Slide 19 text

Summary 19 iPhone12 LiDAR  Introduced Cloud layers function for manual classification  Segment and Merge functions are available for more detailed labeling  Data from the Tokyo Metropolitan Government's Digital Twin Realization Project is used

Slide 20

Slide 20 text

Supplementary material: How to download point clouds 20 iPhone12 LiDAR