3D street level imagery: Securing the missing perspective

3D street level imagery: Securing the missing perspective


Geospatial information is increasingly being displayed in three dimensions. Global players like Google, Nokia, Microsoft and Apple have recently made announcements showing realistic looking 3D maps from a bird”s eye view. But what has been announced so far lacks detail where it matters most: at street level.

Geospatial information is traditionally collected and displayed in two dimensions. However, to successfully manage growth, our increasingly complex society and infrastructures require more detailed spatial information. Here the need for a third dimension comes into place.

Many global players like Google, Nokia, Microsoft, Apple and Amazon have jumped into the cartography pool and currently present realistic-looking 3D maps that let users move around in major metropolitan areas that are rendered in photorealistic detail, but only from a bird”s eye view. However, as soon as one comes close to street level (if this is even allowed), the building façades either resemble ”melted ice cream” or one is directed into a panoramic ”bubble” that shows the surroundings from fixed positions only. What is missing from these 3D maps is a realistic 3D street level component. Detailed 3D street information is the most important component to generate traffic as it provides the most realistic impression of a city, the locations of advertisers and the places of interest for end users.

The DCR (digital cyclorama recorder) camera and processing technology make it possible to collect detailed and accurate 3D panoramic street level imagery of every street, alley and highway. These recording systems can be mounted non-intrusive on almost any vehicle, vessel or train for rapid collection, processing and upload of highly accurate and detailed spatial data for mapping and GIS applications or just simple viewing.

In the past decade, studies were conducted with high precision LiDAR technology in which the surroundings are continuously mapped by laser distance measurement from driving vehicles and transformed into a 3D point cloud. Via automated data fusion they succeeded in precisely matching this high density LiDAR data with their photogrammetrically correct cycloramas. The resulting so-called ”3D depth cycloramas” enable new possibilities e.g. precise overlay of 2D and 3D map data (taking occlusions into account), single click measurements of points, distances and areas, as well as road profiling. Although this automated data fusion is highly feasible, it showed specific errors due to distortions, which are inherent to LiDAR point cloud data. These errors can be detected, but nevertheless lead to unreliable depth data.

As a result, this solution will not give high level quality and reliability. Further, a recording system comprising LiDAR sensors is expensive in terms of investment and operational and maintenance costs. For these reasons, CycloMedia started to investigate alternative techniques, solely based on image content, e.g. Structure from Motion (SfM) and Dense Matching. These techniques are, in principle, based on a process of identifying identical points in a set of successive images. By calculating the 3D location of these points, one can create a 3D point cloud that can be used to fill in the photorealistic 3D street level component that lacked in the bird”s eye view. Apart from a ”simple” 3D point cloud, 3D meshes and textured 3D meshes can also be produced, combined with cosmetic correction of noise, point cloud thinning (for data reduction) and finally the generation of 3D models (containing object descriptions).

Large-scale production of 3D point clouds from their panoramic images is the way forward. Key in this implementation is the precision and reliability of every single point in the 3D point cloud, as well as high throughput enabling large-scale projects.

Work is also on for extremely high resolution and high throughput next generation recording systems, specially designed for global roll-out of its SfM 3D processing pipeline. This focuses on resolution, image quality, positioning quality and recording speed. Together, these developments are aimed to lead to even higher quality and even more precise and realistic 3D street level data for a multitude of applications.