SPOT

• 

  Chair Detection

• 

  Occupied Chair Detection

• 

  Table Detection

Inspiration

Long times of waiting at the airport looking for a seat, or reaching a restaurant or food court and not being able to find a table for 4. It's these sort of situations that made us think that there is a better way to deal with the situation. Thus we came up with the idea of SPOT.

What it does

It is a computer vision solution that is able to detect seats/chairs occupied by a person. Thus helping us to develop a real-time understanding of the occupancy of a seating space such as malls, airports, restaurants, libraries, cafes, and way more. It further provides the user with historical inference to help them plan their visit to such establishments.

How we built it

We made use of Facebook's Detectron library with a custom object inference algorithm to detect the number of available seats viewed through CCTV feeds. We then mapped these CCTV cameras to rooms and developed an iOS app which guided users to unoccupied seats via a virtual map. We made use of a Flask Server which communcated with the FrontEnd and the Machine Learning Platform hosted on GCP. We further made use of AzureML studio to predict waiting time of a table using a boosted decision tree.

Challenges we ran into

Understanding the Machine Learning APIs, and combining them to achieve the desired outcome took significant effort and perseverance. Furthermore, dealing with the connection of the various links to develop the end to end system took quite some time and effort.

Accomplishments that we're proud of

Developing a robust system from scratch, especially within the given time frame is something we are really proud of. We hope that this system proves to be worth the effort we've invested, and help numerous others deal with a daily life issue.

What we learned

We've learned a lot more about working with the CLI and newer frameworks like Flask. Apart from this, we got a deep knowledge of Facebook's Detectron API as well as the AzureML Studio.

What's next for Spot

Extend it for other common seating environments, improve the accuracy and add additional features like a voice assistant and anomaly detection.

Built With

• azure
• boosted-decision-tree
• densepose
• detectron
• flask
• gcp
• ios
• machine-learning
• python
• swift
• uikit