I could just use Street View, right?

Vehicle or street level (i.e. mobile mapping) data comes in many forms, and although they might serve different purposes, from the client perspective, they could appear to do the same thing (i.e. the “why can’t I just use Street View?" question).

The main categories of Mobile Mapping data, in my opinion, are:

Roof Mounted:

• Specialist survey vehicles that capture Lidar and Imagery.
• Camera vehicles that capture high-quality imagery or video.

In-Car:

• Dashcam/mobile phone systems that capture video or imagery.
• Connected Vehicle systems that could be a combination of sensors and cameras. The sensors and cameras are embedded into the car itself.

?

There are essentially two main differences between the data types: accuracy and frequency (or availability). It’s common for people in the industry to talk about “survey grade” or “mapping grade” to describe the different sorts of systems. I quite like how Chris Stretton MSc MIET describes the systems:

• Low Frequency / High Fidelity
• High Frequency / Low Fidelity

?

Usually, a high frequency system will have lower fidelity (quality) but will of course benefit from a much greater volume and availability of data. In theory, the price should be lower too. There are systems at both ends of the spectrum, although I acknowledge that the gap between systems is reducing as connected car sensors become more advanced and the number of survey vehicles increases.

?

Low Frequency / High Fidelity

Specialist survey vehicles are examples of low frequency but high-fidelity systems – customers typically commission a single survey of a specific stretch of road. This data is usually used when planning road changes, or for detailed inspections. This data may be captured only once, or perhaps no more than once a year. Often the data is used extensively for a short period of time before being “put on a shelf”.

These systems include very accurate GPS* receivers and IMUs to ensure that the data inherits strong positional accuracy (usually measured in centimetres).

?

High Frequency / Low Fidelity

Connected Car data is at the other end of the spectrum. Data is captured/created in huge volumes, but the positional accuracy of the data is often poor because it relies on basic GPS receivers.

Positional Accuracy

It’s worth spending a moment to dive a little deeper into positional accuracy and how this applies to Mobile Mapping data. There are two broad ways to consider accuracy* for Mobile Mapping data:

·??????? Camera Position

·??????? Feature Position

Camera Position

The imagery (and Lidar) captured from a vehicle includes only the position of the camera itself. In other words, the data includes the Lat/long of the vehicle – essentially, where the photo was taken from. Any issues with GPS signal (or other sensors) will reduce the accuracy of the position recorded in the data. Most of us are familiar with the pulsating blue dot in Google maps, showing us that our position is not very accurate.

In Mobile Mapping data, these issues can often be seen as a “drift” from what we know to be the probable location of the data (i.e. the centre of the road). There are ways to remove/reduce these issues, but all involve additional sensors or post processing. Most data providers will give you an average accuracy metric – this is usually how far their data is from its true position (often tested using ground-truth data).

Feature Position

This is where things get a little interesting. If we’re using the Mobile Mapping data to create further data (such as a map) then we’re likely to carry out some sort of feature extraction (FE), either manually (via heads-up digitising) or automatically via some sort of Artificial Intelligence (Automated Feature Extraction - AFE). Increasingly, AFE is becoming the standard method for creating maps from Mobile Mapping.

The simplest AFE methodology will merely locate a feature in an image/video and provide the location data extracted from that image (i.e. it will tell you where the camera was, not where the feature is). This might be ok for some use-cases but will usually not be sufficient for more advanced uses. We therefore utilise various algorithms to calculate the location of the feature relative to the camera’s position (with Lidar we don’t need to do this as the Lidar data itself already includes it).

The position of the feature therefore inherits any errors in the camera’s position plus any limitations in the algorithm we’ve used. For example, we might design a system that can identify the same feature in 3 or more photos and then triangulate the feature’s position accordingly. The result would be a map of features in their real-world position but with an accuracy usually measured in metres rather than centimetres.

We should ensure as an industry that we look out for incident where this happens and be confident in challenging customers on their actual needs

There is a common misunderstanding in the world of Mobile Mapping that camera position is the same as a feature position, and sometimes this creeps into a customer’s statement of requirements. We should ensure as an industry that we look out for incident where this happens and be confident in challenging customers on their actual needs (they may have simply copied and pasted the accuracy requirements from an online spec sheet).

?

Mobile Mapping Use Cases

To understand a little more about how we might approach the “fidelity v frequency” question, we can explore potential use cases for Mobile Mapping. I’ve mapped out a few of them in this chart – giving each a score out of 10 for each.

This is of course a simplification of each Use Case, and some will involve sub-tasks that could change the data that’s required. For example, pothole detection would require high frequency but low fidelity data if the data was only used to alert a local authority to the presence and volume of potholes. But if the data was going to be used to estimate the volume of concrete needed to repair potholes, then we’d need much high-fidelity data (but not as often). Each use-case is nuanced and there isn’t a one size fits all, but I’ve tried my best to map each based on the main application.

I'd love to know if you have alternative uses or of course if you have a different view to mine.?

?* I of course mean GNSS, but for the sake of simplicity I am using the commonly understood “generic term” GPS.

**I am using the word accuracy in the broader sense of how a customer might perceive the eventual data/output quality. I won’t get into the differences between quality, accuracy and precision here!

Also, thanks to Dall-E for the wonderfully realistic (!) image of a survey/Lidar vehicle!