A sensor is a sensor is a sensor. If you buy a lidar sensor, you get a mechanical device. Every 10th of a second, it will produce and send a point cloud (a binary mapping of the space with millions of data points). To get usable data, you need two additional software pieces (perception software and people-measurement analytics) – both of which have a dramatic impact on the final quality and usability of your data.

The lidar sensor passes raw point-cloud data to the perception software. The perception software identifies objects in the scene and creates a simple data feed consisting of object identifier, object classification, timestamp, and position (additional fields may include box-size, velocity, and number of points). The people measurement platform ingests that feed, cleans and contextualizes it against a digital mapping of the space, and provides real-time and historical analytics.

A typical system looks like this:

The perception software nearly always runs on a local (on-premises) server. That’s important to understand. Point-clouds are BIG, and they are created 10 times a second. Sending them to the cloud usually isn’t an option.

A single perception server can manage multiple sensors. How many it can handle depends on the density of the point-cloud, the frame rate, the perception software, and the processing power of the server. Usually, the answer is somewhere between 5-12 sensors per processor.

The people-measurement platform, on the other hand, can live on-premises or in the cloud (our DM1 platform supports either deployment model). This works because the feed from the perception software is orders of magnitude smaller than the raw point-cloud.

In the last post, I described point-cloud density as the KEY metric for judging the people-measurement capability of a lidar sensor. But while that’s true, the real-world performance of a sensor isn’t JUST a function of point-cloud density. The point cloud collected by a sensor must be analyzed by software connected to the device (the perception software) and the quality of that software matters quite a bit. Even with a highly detailed point-cloud, poor perception software may impair the quality of tracking. And good perception software may make up for less powerful sensors.

Think about the world of smartphone cameras where sophisticated processing algorithms sometimes (but not always!) make up for small lens sizes or lower mega-pixels. Not only is performance a function of both hardware and software, that function is not constant across different applications. Sometimes you NEED the raw resolution. Sometimes good software can effectively fill in the gaps.

Real-world performance in terms of object identification and tracking is always a combination of hardware and software.

Keep in mind that perception software is not free even when it’s provided by the manufacturer. It’s almost never included in the price of the sensor and costs vary widely. It also requires additional hardware to run on (which may also be expensive or even proprietary). So, when you evaluate the cost of a lidar system, you need to include the perception software you’re going to run and the server you’re going to run it on in addition to the raw sensor cost. That super-cheap made-in-China sensor isn’t much good without perception software to match.

Some lidar manufacturers make their own perception software. Some don’t. If they don’t, then you’ll be using 3rd Party perception software that typically supports multiple lidar types and vendors.

There are pros and cons to each approach. With the manufacturer’s perception software, you’re getting a single point of responsibility (nice) and software that’s optimized to the sensor (also nice). With a 3rd party, you get the ability to use different sensors (nice) and a company that’s entirely focused on software (also nice).

Which is better?

Maybe a clear answer will emerge, but right now there isn’t one. Given the uncertainty in the lidar market, you might think that getting vendor independence by using 3rd Party perception software is a big advantage. And it is. Except that there’s no guarantee that the software provider is going to survive either – and it’s the software you’re integrating with, and which will be painful to change. Until the lidar market shakes out and matures there’s at least some risk with any choice you make.

If perception software takes you from point-cloud to raw data, the biggest (and most important and interesting) piece is what happens next. A people-measurement platform takes you from raw data to application. Perception software will generate a stream of low-level events each having a track id, a timestamp, an object classification and a scene position. That data typically cannot be used as is, it is filled with line-of-sight breakages and has no spatial contextualization.

To be used effectively without months or years of engineering time doing data cleansing and analytic work, you need to add a people-measurement analytics platform. That platform will ingest the raw data from the perception software. It cleans the data, stitches tracks across line-of-sight breaks, contextualizes the data to the space, and provides a range of analytic functions like heat-mapping and pathing. A people-measurement platform will usually also provide configurable real-time feeds and configurable alerting (to SMS, App, or email) that include cleaned event level data and capture key out-of-the-box metrics for things like queue lengths and wait times.

That lets you easily integrate people-measurement analytics into your own applications (including dashboarding and analysis tools like Tableau) and digital signage while getting the advantage of analytics specifically designed for journey measurement and optimization.

Don’t confuse a people-measurement platform with dashboarding. Though a people-measurement platform may include dashboarding, it must do much more than provide a dashboard.

Your time to value with lidar will be dramatically reduced by pairing up all three components in a single system: lidar sensor, perception software and people-measurement analytics platform.

In my next posts, I’ll drill down on each of these components – the perception software and the people measurement platform.