Day 21 of 30 Day AI Challenge – Can Hotels Practically Use Computer Vision?

For those just joining us, every day during this challenge, I’m going to try and do something different or better, using currently available Artificial Intelligence tools.

We’re now 21 days into this challenge and there’s still a veritable treasure trove of AI capabilities to discover! Today I’m going to take on Machine Vision and see how this can be applied practically. Let’s start with the basics.

What is Machine Vision?

Machine Vision, also known as Computer Vision, is a field of artificial intelligence (AI) that focuses on teaching computers to understand and interpret visual information from the world. In other words, it’s all about giving machines the ability to “see” and process images or videos, much like we humans do with our eyes and brains.

You might be thinking, “But how do computers actually see?” Well, Machine Vision algorithms work by analyzing digital images, breaking them down into pixels, and identifying patterns or features within the pixels. This allows computers to recognize objects, detect motion, and even understand human emotions or gestures.

Practical Uses of Machine Vision

Now that we’ve got a basic understanding of Machine Vision, let’s explore some of the cool ways it’s being used in the real world:

1. Facial recognition: You’ve probably used this feature on your smartphone or social media platforms. Machine Vision algorithms can identify people based on their unique facial features, which has applications in security, authentication, and even tagging friends in photos.
2. Autonomous vehicles: Self-driving cars rely heavily on Machine Vision to navigate roads, avoid obstacles, and follow traffic rules. The technology helps vehicles “see” and interpret their surroundings, making decisions on how to respond accordingly.
3. Medical imaging: Machine Vision can help doctors detect diseases and abnormalities in medical images such as X-rays or MRIs, improving diagnostic accuracy and early intervention.
4. Retail: In the retail industry, Machine Vision is used for tasks like barcode scanning, inventory management, and even creating personalized shopping experiences by analyzing customer preferences and behaviors.

Let’s Try A Little Experiment!

So, you’re probably wondering if experimenting with Machine Vision is a little complex and best left to the pros. Yes, probably. In order to create really useful, sophisticated models, you need some pretty smart people, expensive hardware and a very robust, diverse dataset. But all hope isn’t lost for the tinkerers amongst us.

Enter Google’s Teachable Machine – a user-friendly, web-based platform that lets you create custom machine learning models without any coding experience. It’s perfect for those looking to dip their toes into the world of Machine Vision and understand its basics through hands-on experience.

Teachable Machine allows you to train models for image classification, audio recognition, and pose estimation. All you need to do is provide examples for each category you want the model to recognize. Once your model is trained, you can test it with new images or sounds, and see how well it performs.

For instance, you could use Teachable Machine to create a simple model that recognizes different types of fruits or animals. Upload images of apples, bananas, and oranges, then test the model with a new image, and voilà! Watch as the model classifies the fruit correctly (hopefully!). Here’s a great intro video from Google:

Here are a couple of quick experiments I ran myself:

Are People Wearing Masks?

For this test, I downloaded a publicly available dataset to train the model. I sorted the available images into those with people wearing or not wearing masks…then provided different test images the model hadn’t seen to see if it could correctly tell if the person (or people) were wearing masks. This worked pretty well with my initial test, though as you feed the system more complicated iterations, accuracy rates fall with such a small dataset. Still, this shows just how governments and institutions likely trained surveillance systems during the COVID19 pandemic.

Hotel Beds: Well Made, Not So Much…or Dirty?

In this next test, I used another publicly available dataset and sorted select images showing well made hotel beds, some with room for improvement and rooms with dirty beds that hadn’t been made up. This was a bit more complex, but despite the limited data, the system did a pretty good job.

This sort of thing has plenty of practical uses…for example, using the standards for specific brands, you can train the model with images showing the correct setup for amenities, room or F&B setup, etc. and then use that model with an app provided to Housekeeping, F&B Staff or even Brand Standard Inspectors. In non-sensitive areas, with the right security and privacy guardrails, you could potentially also embed the software into surveillance systems and generate alerts centrally as needed.

Other Use Cases In Hotels:

1. Live guest sentiment analysis: Train a Computer Vision model to recognize and classify guests’ emotions based on their facial expressions. This could help you gauge customer satisfaction and identify areas for improvement in real-time. Of course, it goes without saying that you’d need to factor in privacy guardrails and ensure the system wasn’t used for identification or other undesired purposes.
2. Audio-based assistance: Train an audio recognition model to identify common requests or questions from guests, such as asking for directions, requesting room service, or inquiring about checkout times. This could be used to improve response times and automate basic customer service tasks.
3. Pose detection for wellness: If your hotel offers a gym or wellness area, train a model to recognize and classify various exercise poses or yoga positions. This could be used to create an interactive workout guide or offer personalized feedback to guests based on their form and movements.
4. Room cleanliness inspection: As you’ve already seen in the example above, train an image classification model to recognize and classify various levels of room cleanliness or identify specific objects such as towels, toiletries, and bed linens. This could be used to optimize housekeeping processes and ensure consistent room quality.
5. Food identification: Train a model to recognize different types of dishes served at your hotel restaurant. This could be used to create a smart menu or help guests with dietary restrictions identify suitable options.
6. Security and safety: Train an image or pose detection model to identify potential safety hazards, such as unattended bags, suspicious behavior, or people not wearing masks (if required). This could be used to enhance security and safety within the hotel premises.

These are just a few thoughts, though knowing the industry, tech and laws (many still evolving and to come), there are some pretty steep hurdles to practical implementation.

Wrap Up

Machine Vision is an exciting field with a growing number of real-world applications. If you’re inclined to try it, Google’s Teachable Machine offers a fun and accessible way for anyone to experiment with Machine Vision concepts, even without coding expertise. Remember that while Teachable Machine is an excellent tool for prototyping and proof-of-concept projects, it may not provide the most accurate or robust results for large-scale, production-ready applications. It’s essential to evaluate the performance of your models and consider more advanced machine learning tools and techniques if necessary. You also need to factor in privacy, security, local law and other common sense considerations when considering the practical implementation of these systems.

Till tomorrow…