How This AI Company Built Country Scale Maps in 3 Months

The Problem

A leading AI cloud computing company approached us for a bespoke cost-effective mapping solution that could efficiently map the entire UAE in merely 3 months. Their need for a large scale, end-to-end solution with custom map data attributes presented a unique set of problems:

1. Building country scale maps with a fast turnaround time 

Creating high precision, country scale maps with ability for rapid refresh rates is an expensive and extensive endeavor. The process involved:

• Building a map at a UAE level — end-to-end — in three months
• Adding 50+ custom attributes
• Performing quality checks and conflict resolution
• Constantly maintaining and refreshing map data

2. Derive map data intelligence from multiple imagery sources

The street level and private imagery shared by the client was captured in multiple drives. Getting a holistic view of the data and deriving intelligence from it was a time-consuming process. The team had to:

• Constantly shuffle between streams of data captured in different drives to get a complete picture of a location
• Check the data for duplication, inconsistencies and inaccuracies
• Analyze signboards to determine road overlays and other such features
• Study street imageries of road structures and junctions meticulously to create routing tags like road restrictions

3. Easily integrate with routing and navigation engines

The data structures used by different routing and navigation engines like OSRM and others vary by routing engine type. The client needed map data that could easily integrate with their existing routing and navigation engines. The client required a bespoke solution that could:

• Effortlessly integrate with their downstream routing and navigation engines
• Produce map data that could be easily consumed with minimal/no assistance from map solution providers
• Follow the standard schema and tagging structure along with tailored specifications provided by them

Our Approach

We started by analyzing the company’s unique needs, which provided us with nuanced insights into the problems they were facing. These included:

1. Existing map data vendors’ inability to support custom map data layers

Existing map data vendors follow the ESRI / Shapefile standard of tagging and don’t offer customized or hyper-localized solutions based on what the specific customer’s needs are. This prevented this AI company from leveraging valuable proprietary data such as street imagery and GPS data as these can only be integrated into a platform that can support custom map data layers.

2. Impact of poor-quality navigation system on routing efficiency and revenue

Navigation needs high data quality as the navigation attributes directly impact routes. A drop in its quality results in inaccurate routes, which in turn leads to poor customer experience, platform inefficiency, loss in revenue, and potentially safety issues for the drivers. Thus, ensuring high-quality navigation was one of our top priorities.

Our approach was to create custom map data layers with tailored routing tags that can be used by their internal navigation APIs or any internal tools that they support.

Our Solution

A factual query was built to highlight and auto-fix redundant traffic signals across UAE

Our analysis indicated that solving this multi-faceted problem required a specialized approach that’s blended AI with map expertise.

With this, we implemented the following 4-step plan:

1. Built a platform to overlay proprietary imagery data in an integrated map-editing platform

NextBillion.ai recognized that the requirements needed a custom platform with automated tools. We began by consolidating and verifying the proprietary street level imagery as well as private imagery data. We developed a tailored platform that could easily:

• Merge and align proprietary street level imagery, private imagery and open source imagery
• Integrate client’s proprietary imagery data to ensure that their vendors and internal map team could use it with ease
• Incorporate the derived data intelligence into one map stack
• Follow the schema and tagging structure of the client’s internal routing and navigation engines

2. Used AI modules to efficiently extract map attributes from street and satellite imagery

We met the need for high-quality navigation by relying on AI to identify map attributes from street and satellite imagery. For example: Most intersections have traffic signals to regulate the flow of traffic. We used machine learning to detect intersections first and then cross-verified if the traffic signal has been tagged or added twice. These AI modules helped us to:

• Auto detect attributes such as junctions and traffic signals
• Flag duplicates
• Maintain the high-level quality needed of navigation attributes

3. Tasking platform to utilize crowdsourcing for lower cost as well as faster turnaround time

To speed up the process and meet the deadline in a cost-effective manner, we decided to leverage existing OpenStreetMap (OSM) schema. This helped us to save valuable hours and achieve a faster TAT. We used crowdsourced spatial data by:

• First, running automated checks to ensure that the map data quality is up to the desired standard
• Building additional intelligence on top of the OSM schema, and creating a custom layer with the help of customer’s proprietary data
• Checking the correctness of the data using the street imagery that we got from the client

4. Built custom QA tool to perform strict platform checks on attributes to ensure high quality

Due to the built-from-scratch and customized nature of the solution, we designed a special QA tool to auto-detect internal issues. This helped us to avoid manual checks that can be expensive and inaccurate. It allowed us to:

• Build automated scripts that could detect incorrectly marked tags
• Quickly identify, review and rectify any errors
• Write QA checks based on customer’s feedback for selected attributes
• Perform unique checks on all the custom tags and specifications requested by the customer to ensure extremely high data quality and consistency

The Outcome

Attribute-wise project visualization

We created a high-quality mapping solution that was tailored entirely to the customer’s unique needs and was delivered with a TAT of just 3 months.

With our custom mapping solution, the organization succeeded in:

• Mapping over 25,000 miles of roads with 50+ attributes — including 10+ custom attributes not supported by any other existing map vendors
• Saving costs by 70%
• Adding 20% more POIs than requested
• Reducing the turnaround time by 50%
• Achieving a map navigation quality of 99.5%

The centralization of the mapping solution with a single vendor also made the company’s day-to-day operations much simpler.

If you’re interested in building country scale maps that allow for better control, customization, and profitability for your business, reach out to us now!