LiDAR Training Guide

 What this tool's for?

  • Our customers run these types of jobs to train computer vision algorithms to detect and track objects in 3D Point Cloud (LiDAR) scenes. It's a scene that was scanned and stead of a usual camera, its done with a laser beam and the resulta that we get as a cloud of points, the reason of the name. LiDAR means Light Detection and Ranging, but that's not crucial to understand the tool usage. 
  • To better understand it imagine you were trying to build a self driving car or a robot to help keep factory workers safe - it would be really important to understand what type of objects are out there, where they are located and how objects move.
  • In order to make these algorithms successful they need lots of human labeled examples to show the computer and teach (aka “train”) it what to do
  • Customers will tell you what they are trying to understand and track in this Lidar/3D Point Cloud scene by using an “Ontology” - this is just a list of things (called “classes”) that they want you to label in the image
  • Our tool uses uses machine intelligence to help you use your human intelligence to build intelligence - How cool is this?! .

Quick look in tool's functionalities: tool anatomy

A quick view of the screen you will be working with to quickly understand the features.

Some vocabulary before start

  • 3D Point Cloud Annotation Tool
  • This is the tool you’re looking at now.
  •  
  • 3D Point Cloud
  • A point cloud is a set of data points in space.

  • Interpolation
  • This is an algorithm that can predict the route of different objects from frame to frame. 

  • Ontology
  • This is the list of items to annotate in the image. It’s on the left and can be short (1 class) or long (1000 classes)

  • Class
  • A class is a single type of object in the ontology to be annotated. This could be “car”, “pedestrian” or “dog”

  • Frame
  • A single scene from the sequence of scenes

  • Annotate
  • A fancy word for drawing on an image - when you put a box around an object in an image, you “annotate” that object

  • Cuboid
  • The technical word for the shape you will use to annotate

  • Computer Vision Algorithm
  • The type of machine learning (code) that understand and processes 3D Point Cloud data

  • Occluded
  • This is another term for “hidden by something”. If a person steps behind a tree so you can’t see them, they are occluded by the tree.

  • Scrub
  • This means moving between the frames of a scene (back or forth)

The task:

  • Attention: To achieve an optimal use of the tool we suggest using it on full screen and with a mouse. You can do it using a trackpad, but it can be harder.

a) The space: