25/8 2021 ・ Articles / Data insights

New in Annotell Platform: HD Mapping

Lately, the team at Annotell has been working hard on expanding our platform capabilities with an HD Mapping tool. In our quest to make fully self-driving cars a reality we acknowledge the need for high-definition maps. However, building and maintaining these maps is a difficult task, for which currently no good tools exist/that no tool has managed to overcome (so far). That’s why we are set on creating a high quality tool which solves the existing problems with HD-mapping! 

Why are HD Maps important? What challenges need to be overcome? What problems will our solution solve? Read all about it in this article! And if you want to learn more, don’t hesitate to reach out!

Why build HD Maps?

HD Map is a vital part of autonomous driving. HD Map can contain loads of vital information on where cars can and can’t drive, where other traffic participants can and will go, and which areas to pay extra attention to. We believe that harnessing HD Maps effectively makes autonomous vehicles safer by complementing the onboard sensor system. Because the car will know what situations are coming up it can provide enhanced driving comfort and reduce its emissions.

What to label

At Annotell we are able to annotate all the properties on the road which make up sufficient HD maps with centimeter-level accuracy such as:

  • Intersection and lanes
  • Road and lane dividers
  • Driving path, bicycle path
  • Walkways and pedestrian crossings
  • Turn stop and precedence areas
  • Car parks
  • etc.
  • Your specific metadata. Want to know where each bus stop is? Parking areas for schools? The location of each drive-way for food-delivery? We support your custom preference!

What data to use

The Annotell mapping tool supports the following data as input:

  • Lidar top-down projections
  • Satellite images
  • Camera images together with location data
  • Drone data

Data is represented as layers of information to our trained workforce who annotates it. When given the global location of the data Annotell also uses existing maps to update your HD map.

Before HD Mapping

Annotell Platform: Bird's-eye view before HD mapping

After HD Mapping

Annotell Platform: Bird's-eye view after HD mapping

Changing environments

The big problem with mapping the world is that the world is constantly changing. At Annotell we support the following changes in our platform to keep your map up to date in an effective way:

  • Static changes. Whenever a new road opens up, or a new sign gets put up, the base-layer of the map needs to be updated. We expect this to happen on a yearly basis, and treat the map we make here as the basis for all future updates for specific areas.
  • Semi-static changes. Roads undergo construction, sometimes for weeks/months at a time. When this happens we add the construction zone to the map with the new route within the construction zone, so self-driving vehicles can keep operating in that area. As soon as the construction ends we remove this temporal layer so that cars can drive their normal routes again.
  • Semi-dynamic changes. Sometimes routes one has to drive change temporarily. There might be an event, someone might conduct work on a man-hole cover, or a moving company might block part of a road to do loading and unloading. At these moments the Annotell tool can easily block parts of the road for a short period of time.
  • Dynamic changes. When an accident happens part of a road will be blocked off for a short period of time. Use the Annotell tool to mark this down, and changes will be automatically removed after a set period of time.

Supporting change is the core of the Annotell mapping tool, as this is the only way to prevent so-called map-rot. By quickly detecting differences between the real world and the mapped area the map becomes a self-healing part of your autonomous stack.

The Annotell tool supports both manual as well as automatic change detection. Our intelligent algorithms find areas of likely-changes, which are sent to mapping experts who either update your map, or discard the change.

High Accuracy and ADASIS

We are fully aware that high accuracy is the biggest challenge for producing an HD map. Annotell is experienced in producing high quality ground truth for all kinds of autonomous drive related tasks. Our team of experts has deep industry knowledge and know-how to guide the projects and focus on what matters most i.e. to provide the most efficient ground truth for our clients. In the HD map specific domain, our expertise is fully up to date on what is required to fulfil ADASIS protocol. Our platform is secured with the features needed to create the data for building the HD map. Our checker app is safeguarding the quality of the produced data at centimeter-level.

Quality assurance

Centimeter-accurate maps are vital for self-driving vehicles. We therefore invested a lot of time into our quality assurance process. Our ‘checker-app’ helps check in the map we produce to make sure it is of high quality. It contains a set of heuristics which help detect hard-to-catch errors with a simple visual inspection. The checker app ensures centimeter-accurate maps. Before exporting our map we look for common mistakes and irregularitiesto make sure they never reach your vehicle.

Aligning data

In robotics projects, robots drive through a representation of the world, which is their map. When creating HD maps it’s important to align any updates to the old base-map, otherwise each update becomes a costly operation in which one re-annotates the whole map. We align sensor data precisely on your old map using hand-picked landmarks. This way mapping experts can see the true difference between the current state of the world, and your previous map.

Divide and conquer to reach scalability

Precisely mapping an environment is a large and complex task. We believe that by splitting an area into sections and splitting the annotation tasks into different layers, we can efficiently scale a mapping operation across multiple experts. Annotell works with many trained experts, which gives us the ability to scale mapping operations while retaining a high quality map. And the most important enabler for such operations is that Annotell tool is equipped with the features actually making all these splits and merging possible.

As one never knows how many updates are needed on a daily basis we provide a flexible workforce of mapping experts on demand. On days where the world did not change significantly these mapping experts are working on other tasks necessary for autonomous driving. We allow easy ramping up and down of your map update work. This way you always know that your map is updated as fast as possible, in a cost efficient way.

We hope to have convinced you that HD maps are a vital part of any autonomous stack by complementing your onboard sensor system. At Annotell we are set on making self-driving cars a reality, building and updating HD maps in a scalable way is an important part of this. Please reach out to us if you have any questions about HD mapping, or if you are currently experiencing problems with your current HD mapping tools.

Written by

Photo of undefined

Roland Meertens

Product Manager

Photo of undefined

Polly Jing Li

VP Advisory Services