What is Bluetooth's range?

Bluetooth's range for most commonly used devices is about 10 meters, though more powerful devices (or devices that use Bluetooth Low Energy and thus consume less power) can typically send signals up to around 80 meters.

What affect's Bluetooth's range in indoor environments?

Three core considerations impact Bluetooth's range performance indoors - the number and size of obstructions (such as walls) between communicating devices, the power of the devices in question, and the amount of competing signals, which can cause interference and limit Bluetooth range.

MapScale® digitizes and delivers 3D maps faster and more accurately at scale

Technology continues to advance, and as it does so, it opens up new opportunities to further enhance indoor mapping and the digitization of buildings.

In this blog, we interviewed our machine learning expert, Melih Peker, to discuss:

What is MapScale®? - how our patented platform allows us to digitize thousands of floorplans into interactive digital maps at scale for inclusion in wayfinding solutions

Challenges associated with vision-based computer modeling and how we address them
The 4 steps to train MapScale®'s machine learning models and produce beautiful, user-friendly 3D maps
Explore Pointr's vision for the future as we lead in this evolving field

What is MapScale®?

MapScale® is Pointr's patented AI mapping solution that can swiftly convert raw CAD files (floorplans) into interactive, visually appealing 3D digital maps for end users. Its impressive speed and scale are demonstrated by two examples:

Manual mapping of a 20,000 sqft area would take approximately 6 hours (excl. QA), whereas MapScale® accomplishes the task in under 6 minutes (excl. QA), reducing map digitization time by over 90%! Additionally, MapScale® can be configured to run thousands of floors in parallel, speed which has enabled a major US department store to digitize over 2,000 store maps every Sunday.

How does MapScale® use visual machine learning?

When MapScale® fetches an AutoCAD drawing, it uses machine learning algorithms we developed to identify fixtures such as furniture and walls - just by looking at the image data provided in the AutoCAD drawing.

Additionally, we've created modules that can identify text annotations using data such as the names of the rooms and other text to identify the functionalities of the rooms and furniture on the floorplans.

Two of the biggest challenges when going through the automatic mapping process are that AutoCAD files with diverse designs from different architects include unnecessary elements like hand drawings, writing, version logs, voids, lights, power lines, water pipes, and more.

Examples of these challenges include:

Different buildings, different architects/designers

Lots of irrelevant information, sketches, text, and legends

Despite these challenges, MapScale® can intelligently determine what should be included or excluded, adapting to different floorplans without the need for manual modifications for each customer. This flexibility allows us to offer a customer-agnostic solution that can scale infinitely, making MapScale® a versatile and efficient tool for producing 3D maps.

Vision-based computer modeling is demanding

To deliver MapScale®, we had certain challenges to meet, such as:

Speed and scalability - we needed to be able to map quickly; taking months or weeks to produce a single map simply wouldn’t cut it and was not scalable.
Complex models - dealing with the complex models produced in vision-based computer modeling is challenging. For example, a single, seemingly simple item, such as a chair, can have many different shapes, sizes, and colors.

Examples of chairs in different floorplans

Huge data sets - models that combine visual information (the image of the lines), vector information (mathematical information such as coordinates, slopes, etc. of the lines of the CAD file), and metadata (text) consist of hundreds of millions of parameters, and to train our complex ML models we needed to feed it with a huge data set of maps.

How we tackled these challenges

To address the speed and scalability challenges, we began by equipping ourselves with the latest and most powerful hardware available. Because MapScale® employs vision-based computer modeling, the graphics card (GPU: Graphics Processing Unit) we chose had to be top of the range and powerful enough to map big venues accurately, quickly, and at scale to train our complex machine learning models. We chose the RTX A5500 GPU, and we used two GPUs working in tandem. The RTX A5500 GPU is commonly used in architectural applications, such as virtual reality walkthroughs of buildings, that require large amounts of memory and lightning-fast rendering of 3D images.

Furthermore, we were also able to train and maintain our complex machine-learning models with large amounts of data that Pointr has gathered over the years, further enhancing their accuracy. With thousands of buildings already digitized, each new piece of information acquired is stored in our database, continuously improving and optimizing our machine-learning modules. Consequently, MapScale® can now effortlessly recognize rooms and furniture using image and text data, thanks to constantly evolving and refining its AI-powered algorithms.

4 steps to train MapScale® machine learning models

MapScale® automates the long and complicated mapping process using our machine learning algorithms. The process can be summarized in the following 4 steps:

Step 1 - extract information from drawing files

Information is extracted from AutoCAD native DWG, DXF, or vectored PDF files.

Step 2 - pre-processing

Removing extraneous information

Unnecessary elements such as hand drawings, writing, version logs, voids, lights, power lines, water pipes, text, and hatching are removed from the CAD floorplan.

Text, hatches, and extraneous information are removed from the CAD floorplan in the pre-processing step

Block references & polylines

Block references include desks, chairs, tables, etc., and polylines are standardized, while undesired attributes and entities, e.g., text, hatch, hand drawings, rulers, and grids, are eliminated.

Removing undesired attributes and entities

Training - requires powerful GPU and large amounts of memory

During training, images are annotated, and these annotations are used to train our machine-learning models. These models are then tested using "unseen cases" to ensure accuracy in identifying objects and other structures in drawings.

Step 3 - analyze and convert

MapScale® identifies floorplan structures, such as building outlines, room boundaries, and furniture, and classifies them based on size and type. The map is converted to a vector format in GeoJSON, universally compatible with most mapping providers, and enhanced with a cartographic design.

Step 4 - efficient 3D map production

In this step, user-friendly 3D maps are efficiently produced on a large scale, ensuring fast and accessible access for users.

MapScale® in action

Here’s a short (2-minute) YouTube video that shows MapScale® in action, as demonstrated by our Machine Learning Engineer, Melih Peker.

Pointr continues to lead the way in providing cutting-edge technology to fuel indoor location services for smart buildings. Our vision is to deliver an end-to-end solution, automating the entire process from CAD input to user-friendly mapping and wayfinding. Ultimately, you’ll be able to “flip a switch” to enable smart buildings quickly and at scale!

Our future vision: smart buildings, simplified

“I believe data-driven approaches are the future of scalable indoor mapping. Every time MapScale® digitizes an indoor map, it learns new concepts from each venue. With this extensive, accumulated knowledge and data, MapScale® will be the ultimate one-stop mapping solution in the near future.”

Melih Peker, Machine Learning Engineer at Pointr

Summing it all up

In summary, MapScale® utilizes a vision-based machine learning model to identify building outlines and fixtures in raw AutoCAD drawings based on its accumulated knowledge. With the support of powerful hardware like NVIDIA's GPU, we have achieved the following advancements:

Improved scalability in data processing
Reduced processing times
Capability to handle complex datasets
Enhanced accuracy in identifying annotations (text-based data)
Potential for developing more powerful training modules, including enhanced baseline models and auto-georeferencing

Need more information?

Learn more about MapScale®'s functionality, features, benefits, and why you might consider switching to the industry's leading technology for digitizing buildings by visiting our on-demand MapScale® product webinar.

You might be also interested in: