Computer Vision meets … vomit?!

#40yearchallenge … Steven Sasson, inventor of the first self-contained digital camera (l) and detecting vomit using Computer Vision (r).

Sometimes you start a career and have no idea where it will take you. That’s certainly been the experience of Cleanride’s engineers, who this week began the challenging (and disgusting?) task of detecting vomit using Computer Vision and Machine Learning.

Cleanride is an AI-powered solution for identifying issues in shared vehicles. As car sharing and other new forms of mobility become increasingly popular, dirty & messy vehicle cabins are becoming more and more of a problem.

To inform operators about issues in real time, Cleanride uses cameras mounted in shared vehicles and a YOLO-like model for object detection. This means that the team has to train the system with hundreds of annotated images. We also use CNN-based neural network models to determine if parts of the vehicle cabin are covered by objects that shouldn’t be there.

This includes spills — which is where the vomit comes in.

Second prototype

After presenting our first demonstration at the VW Nutzfahrzeuge Innovation Challenge 2018, the Cleanride team has been hard at work on a second, live prototype to show to potential partners.

The first live demonstration of Cleanride as finalists in the VW Nutzfahrzeuge Innovation Challenge 2018.

Whereas previously the team demonstrated our ability to detect objects like phones, wallets and keys (Level 1 detection), our focus is currently on expanding the types and number of objects. We are also adjusting the neural network structure to gain better accuracy and experimenting with different camera sensitivities and light conditions to make spills (Level 2) more detectable.

Level 1 — Objects such as phones, wallet, keys and rubbish (eg food wrappers, bottles, food scraps, dog faeces) and Level 2 — Liquid spills (including vomit).

Thankfully for the upholstery in the car we’re using for prototyping, the vomit is fake — for now. Meaning that we have some interesting purchases (and recommendations) in our Amazon histories!

An interactive 360 view of a car interior as part of our experimentation with camera placement.

Proof-of-Concept

After demonstrating our expanded Level 1 & 2 capability to partners in February and March, our plan is to begin a 12-month Proof-of-Concept stage focused on achieving Level 3 detection.

This means the ability to identify dirt and smaller particles (mud, sand, dog hair), in combination with odour/particles sensors, and the ability to assess using AI whether the vehicle needs to be cleaned.

Level 3 — Dirt and smaller particles (mud, sand, dog hair), in combination with odour/particle sensors.

UPDATE 31.01.2019: We’re moving on to some pretty funky simulated liquids. Goodbye upholstery!

Sorry if you’re eating lunch right now.

Learn more

There has been a lot of interest in Cleanride recently and we are very encouraged by positive feedback on our progress to date.

The Cleanride solution has the potential to be applied to many different environments and use cases, beyond just vehicle sharing. These include facilities management, monitoring of public spaces and many other safety & security situations.

If you’d like to learn more about Cleanride and the team’s methodology, or are interested in supporting us in a Proof-of-Concept or pilot, please send us an email: hello@cleanride.io and we’ll get back to you.

You can also follow our progress on Twitter.

Thanks for reading!