What are the Limitations of SLAM Systems?
March 24, 2017
Kudan AR tracking
April 25, 2017
Show all

Kudan SLAM and KudanCV – Roadmap update

We often receive questions from our users about our work and our future plans. Find out more about our Unity Plugin, KudanCV, and SLAM here.

What is the difference between the Unity plugin and KudanCV:

One important thing to clarify up front is that our public offering with respect to Unity doesn't reflect what we have available internally.

Our main technology is KudanCV, which as you can guess from the name, contains all our computer vision solutions. One of the main design goals of KudanCV is to provide ultimate flexibility with regard to integration and use-case. If you were to look at the API, you would see that it makes no assumptions about input peripherals, or how the data gets interpreted. It simply takes image data acquired from somewhere and outputs what it can see. This lends itself very well to integration with different kinds of rendering (any rendering engine would be usable here) in the case of AR, but even that isn't a requirement. The data is just as useful to things like machine learning, even if there's no intention of any rendering.

The main goal of development is that we build on the functionality of KudanCV, but because of the heavy links to AR, we take KudanCV and wrap it with KudanAR (which includes our own lightweight rendering) and the Unity plugins, since both of these are common use-cases from our developers and it saves them the trouble of having to integrate. These, however, are secondary to KudanCV in terms of priority, but it's actually pretty straightforward for us to bring new KudanCV functionality to them when required.

Does Kudan support recognition of the following:
- Cylinders (eg. Soup)?

A cylinder and cuboid tracker exist internally but require tools in order to create them as targets. Performance is very similar to the image tracker.

- Conic Sections (eg. Yogurt)?

This is actually a variation of cylinder tracking, mostly around the tooling to create the target.

- CAD-Files?

We don't currently support this but have plans to support registering 3D models in an artist format (eg. FBX, Maya, Blender etc.). All of our tracking technologies rely on targets having a reasonable amount of feature points. We don't intend to do edge-based tracking for objects lacking texture.

- 3D Scanned Objects?


Like everything else, it requires a user-friendly tool to perform the scanning which is why it hasn't been released yet.

- Terrains?

The main thing we work on now is our SLAM, which covers tracking of terrain:

(and many more on our channel).

Our SLAM has been designed for many use-cases. In the spirit of KudanCV, we don't want to make many assumptions about the hardware, so it's been designed to scale across all different hardware configurations, ranging from mobile with the built-in cameras, through to custom vision chips with fancier cameras (stereo, RGB-D etc.).

We have multiple use-cases for it:

  • 1. Headtracking for portable AR and VR devices
  • 2. Mobile AR
  • 3. Robot/drone navigation
  • 4. Additional data for machine learning/AI systems.
What is the vision performance of the image tracker?

Here's a few videos exploring the limits of performance of the image tracker:

This should cover things like angles, lighting, blur. The same underlying tracking technology applies to everything we do so there's similar behaviour.

Can your technology apply to any market segments such as retail, healthcare, industrial maintenance and education? Do you have any “real world” issues?

We try our best to avoid deliberate limitations ourselves but can't always guess them up-front. However, we don't have any issue with adapting when required.

Do you have Hololens up and running in general or just for a limited set of special cases?

We made the decision to not proactively support HoloLens simply because it would be one additional platform we need to maintain (and we already support a lot!). However, KudanCV has no technical incompatibility with HoloLens due to its hardware agnostic design philosophy. Hookup to Unity is a different matter but not for any Kudan specific reasons, just that it would require modification of the plugin in order to access the HoloLens camera, and have the appropriate calibrations.

Can we build tools to automate marker creation?

We will soon release an API for trackable set creation, which means customised tools can be created. You could, then, for example, have a web interface if you wanted, or integrate it directly into Unity.

Comments are closed.