- Voxels might be the best option to build an accurate map of the world
- Not only can space be represented, but also time
- Voxels can create virtual occupancy grid (VOG) maps
Voxels have been around for a long time and they are essentially 3D pixels that can be represented in various dimensions and scales. When voxels are incorporated with a time element, that is voxels can be displayed based on a given time sequence of interest, then it is possible to create 4D maps that incorporate space and time.
While the concepts are not new, creating such maps for the entire planet until now has been out of reach because of the data quantity and computational power required to render such data. In this geospatial podcast, the CEO of VoxelMaps, Peter Atalla, describes how they are creating a global 4D map that can capture outdoor and indoor spaces at different spatial and temporal scales.
Benefits of Voxel-based Maps
We are probably familiar with voxels in games such as Minecraft or augmented reality worlds. However, for mapping purposes, voxels have been relatively underutilised. Voxels can be used to create virtual occupancy grid (VOG) maps.
The benefit of voxel-based maps is they can provide accurate volumetric measurements. By making voxels multi-dimensional, that is multi-voxels in spatial scale, then it is possible to create maps that can represent broad landscapes or interior of small spaces. With cloud-based computing and storage technologies, the vast amount of data required to create and render maps at different temporal scales is also possible, enabling VOG maps to be 4D.
Challenges to Creating an Accurate Global Map
A grand challenge has been to create an accurate global map that anyone can use and represent not only outdoor but also indoor spaces. VoxelMaps is now using cameras, GPS devices, hyper-spectral images, and sensors that can be carried in backpacks or mounted on vehicles to capture data around every corner of the world. Such data can also be used to identify what surfaces are.
The use of deep learning neural network models can also automatically segment and classify features to distinguish landscapes, buildings, and other features captured in imagery. Currently, over 200 types of objects in urban environments can be automatically segmented by in-built neural networks.
More trained extraction capabilities are added so that maps can be built more easily as new data are obtained. Data, therefore, are not only captured but individual features from small- to large-scales can be identified and automatically tagged for what they are, building feature typologies.
Additionally, users could also volunteer data, which can be useful in difficult to reach regions and in indoor spaces. The data are all stored in the cloud, while automated deep neural networks segment data as they come in. Extracted features can also be measured very accurately, with the aim being to create a global map that can be as accurate at 1 cm resolution. While this resolution is not possible everywhere, the voxel-based maps can also scale to different resolutions depending on what the data provide.
Another benefit of this 4D map is you can extract sub-sets of the data into more familiar formats such as .nds or shapefiles.
While individual users who want to map parts or areas of the globe would be interested in such a model, the intent is to make these models useable by machines. In particular, robots, autonomous vehicles, UAVs, and devices such as IoT components.
Vehicles can use the ready data without having to depend on GPS potentially to know where they are. By creating such accurate maps at scales down to 1 cm, this means global maps would allow navigation for devices that require very high resolutions.
This also opens up an economic potential for many industries that now can utilise detailed 4D maps. Companies using landscape-scale data, such as for navigating autonomous vehicles, to realtors selling homes and allowing virtual tours of spaces could potentially benefit from a high resolution global map.
Another area of benefit is the development of 5G data. As companies plan where to put their 5G signalling towers, then having an accurate map that can determine potential interference with signals prior to building signal towers can greatly minimize costs and planning, including removing the need for survey. What is also of great benefit is that the data are also being constantly cleaned. The neural network models and processing algorithms are also used to remove redundancies or data interfering with an image captured used to create a map scene. What the user gets is a clean map with the data they expect.
While VoxelMaps is now beginning to make a voxel-based 4D map of the world, we are still not at the point where the data can represent everything we may want to see. The data still have a long way to go to be captured and the company will likely have to expand a wider audience, including volunteers, to collect and contribute data. However, the road to virtualizing the world and highly accurate 4D maps that represent time and different spatial scales are now being built. In a few years, many places could have highly accurate representation using such voxel-based maps.
 For more on voxels and their benefits for accurate volumetric representation, see: Javidi, B., Okano, F., Son, J.-Y. (Eds.), 2009. Three-dimensional imaging, visualization, and display. Springer, New York, NY..