AI indoor navigation system

Our client – a company that specializes in providing indoor mapping and navigation, no matter the premises. They needed to build an infrastructure for accurate indoor mapping and positioning to support rising demand.

As the project progressed, our team was extended to provide DevOps practices and legacy code refactoring. Moving further over the project’s course, we provided MLOps practices to deliver a quality control solution, which allowed our engineers to evaluate the accuracy of the inference service.

The project’s key objectives:

Achieve optimized response times and costs

By implementing event-driven infrastructure, we aimed at optimizing:

• Both the response times by ensuring efficient allocation of directed computational resources
• And costs by allowing the system to scale on demand and avoid unnecessary consumption during lower activity periods
• Enable faster software development and release

By implementing DevOps practices and legacy code refactoring for faster software development and releases, we focused on providing:

• Competitive advantage
• Meeting evolving customer demands
• Shorter time-to-market
• Greater maintainability and adaptivity to changes

The service is designed to enable any customer guide visitors through destinations and interact with spaces, benefiting leaders across transportation, retail, recreation, and many other domains.

By providing visual and audio directions for accessibility, the solution is facilitating autonomous navigation. What’s more, the system is driving user experience by enabling great personalization to provide the visitors with information about the entered venue.

The service is user-friendly and easy-to-setup:

• The user is assisted by the company’s technician to scan the location and capture required data
• After scanning, the data is stitched to create a map
• The user can manage the map to control the permissions to updating the location and decide on how future visitors can interact with it
• After finalization, the visitors can navigate the location using both visual and audio guides in real-time

The service is made up of:

1. A custom web application for the client’s internal tech personnel for adding new buildings and points of interest to the navigation system
2. A custom mobile application for the end user to take interior pictures and send those images to the inference service, so that the system can navigate the user to the selected destination
3. And the inference service that extracts specific features from images to match those features with the scanned model of the added building or point of interest – the spot with the best match is used as the predicted location

In particular:

• In-depth research of models considering performance and accuracy
• Event-driven architecture for optimized response times and costs
• DevOps practices and legacy code refactoring
• MLOps services for quality control solution

During the project’s scope, we faced multiple challenges associated with:

The improvement of the SLAM system

The size and similarity of spaces within some larger buildings has burdened image processing and comparison. To overcome the challenge, our engineers autotuned thresholds for each pipeline microservice and minimized potential false positive responses.

The hosting of the SLAM system

Such solutions require significant computational resources for adequate response times at an acceptable cost. To solve this challenge, our engineers went with message brokers and divided the components into individual pipeline microservices that accommodate the required computational resources to meet those needs.

The scaling of the delivered solution

In order to support the scanning of different facility types all across the globe, including airports, train stations, schools, museums, and others, we delivered a custom Infrastructure-as-a-code system, which enables a quick, one-click setup of buildings and clusters.

The monitoring of performance, user behavior, and scaling

To ensure high uptime and facilitate user experience, we established individual monitoring within clusters.

The client was aimed at replacing an existing inference service to accelerate user satisfaction and loyalty, concurrently optimize response times and costs, as well as streamline localization precision.

By leveraging the upgrade, the company can harness:

• New customers, higher demand, and expanded revenue streams
  The client can attract more customers and leverage future opportunities for growth and profit.
• Business agility
  The client can introduce new features and updates more frequently, invest strategically, and adapt to changes without compromising business performance and reliability.