What's the point of manually classifying point clouds?

Discuss all aspects of Pointly here.
3D point cloud classification made smart, fast and easy.
Post Reply
User avatar
Sid Pointly
I have made 20-30 posts
I have made 20-30 posts
Posts: 21
Joined: Mon May 11, 2020 4:32 pm
3
Full Name: Sid Hinrichs
Company Details: Pointly GmbH
Company Position Title: Sales and Strategy and Finance
Country: Germany
Linkedin Profile: Yes
Location: Germany
Has thanked: 2 times
Been thanked: 6 times
Contact:

What's the point of manually classifying point clouds?

Post by Sid Pointly »

Hi everyone :)

this post if for everyone who is not familiar with manual classification of 3D data and what it's needed for.

May be let's start first, what manual classification means: It means to assign every point within the point cloud to a respective object class. This you can see in the short illustration down below, where the classes "Roof" and "Building" are assigned to their corresponding points:
manual classification with segment selector in Pointly.gif
Currently there is a lot of inefficient work with point clouds in order to extract information. Pointly offers a way to change that into efficient work. But let’s have a look at two examples to point out the dilemma first:

First example: You manually work with point clouds and do repetitive tasks, which could be automated, e.g. draw within them to extract CAD models. No matter how often you will draw and mark similar objects like a road, you will always have to do it again. The reason is simple: Your used point cloud (data) stays dumb. That means you don’t generate any training data as you are not giving those selected points a sense (a label) while drawing. You could compare it to tracing a picture, where you simply copy information onto another format. Hence, you can’t train a neural network to automate processes like identifying a street and its boarders.

Second example: You use automated classification tools but those are limited, not designed to be scalable and don’t adapt to different scenarios. So even though you might be able to classify e.g. a ground automatically for a DEM, that doesn’t mean you can use that for deep learning and to extract other objects of interest.


So, what’s a way to work more efficiently?


You need to manually classify the point clouds again or start doing it. Even if that feels like taking a step back, it is actually two steps forward. The idea is that you do it once and then rarely ever again. You might ask yourself now “Is this not still a lot of work”? Don’t worry with the help of Pointlys intelligent selection tools the whole classification process will be much easier. Moreover, you can upload, manage and view huge point clouds in our platform.

If you do manually classify, it will greatly benefit you in the future as you can generate training data and faster close the gap between human labeling and machine learning.

Manual Labeling and Machine Learning.PNG
You will get better results, can do automated analyses, isolate objects and complete various other tasks, for example automatically generate lines for CAD models afterwards. Thus, you can train a neural network, which automates and accelerates processes like identifying road signs for the n-th time.

The ultimate goal of manual classification is to gather a decent amount of training data in good quality to train the AI.

If you need help with your AI case we can help you with our Pointly Services: https://pointly.ai/pointly-3d-point-cloud-services/ where we offer things like automated classifications, CAD generation, change detection, automated tree inventories and more :)

Visit us at www.pointly.ai
You do not have the required permissions to view the files attached to this post.
Best Regards
Sid Hinrichs
Pointly
www.pointly.ai
[email protected]
Post Reply

Return to “POINTLY”