Security doesn’t need more data, it needs to make better use of the data already being captured argues Jerone Floor, Senior Director of Product at Opsys Technologies.
Security systems are expected to do something deceptively simple: detect a genuine threat and tell an operator what is happening.
In practice, that is much harder than it sounds. Many traditional systems are good at detecting that something happened, but much less effective at determining what actually happened.
A beam is broken. Motion is detected. Pixels change in part of an image. Each of these can indicate a real intrusion, but they can also be caused by animals, debris, weather, lighting changes or other harmless activity.
Over time, false and nuisance alarms become more than an inconvenience. They consume manpower, slow response and contribute to alarm fatigue.
The problem is not simply that these systems need better software or more tuning. In many cases, the limitation starts with the way the environment is being sensed.
A perimeter, yard or restricted area exists in three dimensions. Objects have size, distance, height, direction and speed. Yet many security technologies observe that space through a two- or even one-dimensional model.
Cameras illustrate the problem clearly. A conventional camera captures a two-dimensional image of a three-dimensional scene. Software can estimate distance, size and movement using perspective, motion and trained models, but estimation is still not the same as measurement.
Beam-based systems simplify the environment even further: a line or virtual plane has either been crossed or it has not.
Both approaches can be effective, particularly when combined with analytics and other sensors. But they share a fundamental limitation.
They are being asked to understand activity within a physical volume without directly measuring that volume. It is time for the security industry to secure three-dimensional spaces with three-dimensional sensing.
LiDAR does this by directly measuring distance and position in three dimensions. It can determine where an object is in space, how it is moving and how it interacts with a defined security zone.
That provides a much stronger foundation for intrusion detection. But it does not, by itself, guarantee a better security system.
As LiDAR matured, however, much of the LiDAR industry began competing on something else entirely: how much 3D data a sensor could produce.
Competition increasingly focused on specifications that were easy to compare and easy to market. More beams, more points and wider fields of view became the most visible measures of performance.
Multiple sensors were fused together to create large, continuous 3D views of an entire site. This made for impressive marketing pictures.
The problem is that better-looking data does not automatically create a better security system. Dense point clouds can be useful, but every additional point has to be transmitted and processed.
Multi-sensor fusion adds synchronization and calibration requirements, while external processing introduces servers, GPUs and additional network infrastructure.
What looks impressive in a demonstration can become much harder to justify once the same architecture has to scale across a real facility.
I have worked with large, fused LiDAR systems where this became a very practical problem.
I have seen deployments require gigabits per second of network bandwidth and high-end graphics hardware simply to transport and visualize the resulting fused point cloud.
What worked impressively in a proof of concept became difficult and costly to deploy at scale.
That gap between demonstration and deployment is easy to underestimate. A handful of sensors connected to powerful compute in a controlled environment is one thing.
Dozens of sensors sharing an existing security network, spread across a live site and expected to operate reliably for years, is something very different.
The issue is not that high resolution or sensor fusion are inherently wrong. They are valuable when the application requires them.
The mistake is treating more data as a measure of better performance. For a security operator, the objective was never to produce the most impressive point cloud.
It was to know whether a genuine security event had occurred, where it happened and what was happening next.
The next step is not necessarily to generate more 3D data, but to make better use of the data already being captured. For intrusion detection, the useful output is not a continuous stream of raw coordinates.
It is an understanding of what is in the protected space, where it is, how it is moving and whether that behavior requires action.
That means shifting from data-centric architectures toward systems that can interpret the scene closer to the point of detection.
On-sensor intelligence can combine direct 3D measurement with tracking, classification and zone logic to convert raw measurements into tracked objects, events and alerts.
Instead of sending everything upstream for interpretation, the sensor can determine locally that a person has entered a restricted zone, where they are and which direction they are moving.
This does not mean raw point-cloud data has no value. There are applications where detailed 3D data or multi-sensor fusion is necessary.
But for many security deployments, continuously transporting and processing every available point adds complexity without improving the decision the operator ultimately needs to make.
The more useful question is therefore not, “How much data does the sensor produce?” It is, “How much useful information does the system produce?”
The LiDAR industry can learn from the evolution of video surveillance. Cameras once depended heavily on centralized processing but increasingly moved analytics to the edge.
Detection, classification and event generation now routinely happen inside the camera, reducing bandwidth, infrastructure and power requirements. LiDAR should follow the same path.
External processing does not only mean a server room or cloud platform. In many deployments, it means installing a separate computer at or near each sensor to interpret the point cloud.
That computer is another device to purchase, mount, configure and maintain. It requires power, adds another network dependency and creates another potential point of failure.
When detection, tracking and classification happen inside the LiDAR sensor, much of that additional processing infrastructure can disappear.
Instead of continuously transmitting dense point clouds, the sensor can communicate tracked objects, events and alerts using a fraction of the bandwidth.
Lower processing and communications requirements also reduce power consumption. That becomes particularly important for remote sites, where every watt affects the size and cost of solar panels, batteries and supporting infrastructure.
None of these characteristics are as visually impressive as a dense point cloud. But they are often what determines whether a system can scale beyond a demonstration.
Security technology should ultimately be judged by what it takes to deploy and operate reliably, not by how much raw data it can generate. The point cloud is a means to that end. It should never become the end itself.
This article was originally published in the September edition of Security Journal Americas. To read your FREE digital edition, click here.