Contact Us

Computer Vision for People Counting: Why it’s the best method

Computer Vision for People Counting

19 April 2023

Blogs

Did you know that using Computer Vision for people counting is one of the best ways to generate actionable intelligence about pedestrian, commuter or customer behavior within physical spaces?

The application of Machine Learning Models and Computer Vision Software – the ability to programmatically detect, analyze and act on predefined objects and actions in frames of video footage – often conjures images of the hyper-futuristic. And yes, sometimes Skynet-esque doomsday visions too.

Skynet

But, as Gartner’s VP of Research, Pieter den Hamer, recently told The Washington Post, emerging Artificial Intelligence (AI) technologies are far more likely to augment existing, monotonous tasks – rather than take-over your job, much less the world.

In the article, AI isn’t yet going to take your job — but you may have to work with it, den Hamer explains that, while “every job will be impacted by AI”, the main overarching benefit to embracing AI is freeing-up humans to conduct less mundane and repetitive work.

So while people counting techniques have existed for a long time, legacy approaches have made it a much slower, more expensive and inaccurate process than it should be. But Computer Vision, also known as Video Analytics, is changing that – fast.

Access People Counting Whitepaper

People Counting of the Past:

Legacy approaches and drawbacks

When it comes down to it, lack of appropriate technology has hamstrung the ability to – and benefits of – logging and analyzing pedestrian data accurately, efficiently and cost-effectively. To understand why a centralized Computer Vision approach addresses these issues, it’s useful to recognize the prevailing deficiencies of alternative methods (such as manual, thermal imaging, wifi, bluetooth, Infrared and laser beam counting).

Common conundrums

While each method of people counting has specific strengths and weaknesses, manual, hardware-centric, server-side and single-purpose techniques of the past share some common challenges:

Scalable video analytics platform

Time-consuming and Unscalable

Single-use-case and server-side approaches require significant set-up time, are often non-repeatable or cannot scale.
Increaser Revenue Generation

Costly to Deploy and Maintain

Physical and hardware-reliant solutions often equate to high up-front implementation costs and expensive ongoing maintenance bills.
Inaccurate Results

Inaccurate Results

Manual and sensor-based techniques struggle in certain contexts (crowded spaces, network interference or fast moving objects) and are hard to optimize to accommodate those circumstances.
Inconsistent results

Inconsistent Results

Older technologies often produce inconsistent results from site-to-site, obtaining dissimilar counts across different physical environments, digital networks or camera types.
Insufficient data

Limited Data Produced for Analysis

Sensor-based and manual methods of people counting capture and create limited data, inhibiting the ability to conduct advanced analytics, which produce valuable insights into human behavior, such as dwell-time or movement patterns.
Delayed processing

Delayed Processing

Many methods are constrained by offline processing, manual collation or limited processing power. Inability to deliver real-time insights also prevents real-time action in security, crowd management or OH&S settings.
Privacy Concerns

Privacy Concerns

Sensor and signal-based methods can reveal sensitive information about individuals, such as facial recognition, location, movements and device data. Decentralized production and processing of video metadata can also lead to data breaches or unauthorized access.
Inflexible

Inflexible and Unadaptable

Built for specific purposes in narrow conditions, hardware-reliant solutions are often inflexible to new contexts and non-applicable to other use cases.
Non-integratable

Non-integratable

Senor-based, camera-specific and manual techniques are difficult to integrate with other systems, inhibiting the ability to generate cross-system insights or a holistic in-context view of people’s behaviors.
Invasive

Invasive

Physical types of people counting – such as manual and sensor-based methods – are often highly visible, can impede pedestrian movement and cause unease due to the obvious observation.
Hard to use

Technologically Prohibitive

Extracting and creating broadly digestible, shareable and actionable insights / data from legacy people counting practices can be difficult and dull.
Sensitivity to Environmental Factors

Environmental Sensitivity

Hardware-based solutions – particularly sensors – are often fixed location technology, which operate optimally in a narrow band of conditions. High temperature, obstructions, reflections, fluctuations in lighting, and sub-optimal distancing can all negatively impact detections.
Access People Counting Whitepaper

Using Computer Vision for People Counting:

Harnessing Machine Learning to analyze video footage

Computer Vision unlocks the potential to digitally analyze occurrences in the physical world in new, productivity enhancing ways. Using Video Analytics technology to count people facilitates more reliable, replicable data collection at scale.

Why Computer Vision is a superior method for counting people

This next section outlines the benefits of computer-vision-based people counting compared to other common methods.

Superior Accuracy

Superior Count Accuracy

Once optimized with relevant training data, Machine Learning models for people counting produce more accurate results than traditional methods, while avoiding the human error and subjectivity associated with people-reliant processes.

Improve Response Rates & times

Faster Processing

Whether post-processing in batches at the edge – or producing real-time results in the cloud – Computer Vision algorithms process video streams with superior efficiency, and make live detections and analysis possible.
Scalability

Improved Scalability

Once trained, Computer Vision models can be applied to many scenarios, scaled across many video sources and large volumes of footage – all without acquiring additional personnel and hardware.
Efficient Set-up and Monitoring

Efficient Set-up and Monitoring

Select cameras, streams or repositories with an appropriate field-of-view, and away you go. Unlike manual, server-side, or single-use-case approaches, fresh set-up and installs aren’t required for each new site. Deployment health-checks can also be performed remotely and programmatically.
More Consistent Cross-site Results

More Consistent Cross-site Results

Because the same people counting models can be deployed throughout a range of locations and video sources, more consistent detections are produced across different sites. Better standardization also makes it easier to compare results between different systems and locations.

Greater Opportunity to Conduct Analytics

Greater Opportunity to Conduct Analytics

Logging detections with Computer Vision enables additional insights beyond counts – from analyzing movement patterns, busy times of day, or congestion points by location.
Enhanced flexibility

More Efficient Adaptation

Unlike people counting methods that rely on physical resources, Computer Vision algorithms can be quickly trained and optimized for different environments – from indoor and outdoor spaces, varied lighting scenarios, and challenging weather conditions.
More Integratable

More Integratable

Unlike manual and sensor-based people counting processes, Computer Vision algorithms and metadata outputs can be embedded and sent to cameras, control systems, Video Management Systems, centralized databases and analytics solutions for further analysis.
More Cost Effective

More Cost Effective

Hardware- and human-reliant methods of people counting are resource intensive. Computer Vision empowers organizations to conduct counts – and deeper analysis – far more efficiently. Easy set-up and remote maintenance, scalability, adaptability and integration reduce costs significantly.
Access People Counting Whitepaper

But, not all Computer Vision for People Counting solutions are created equal

Generally speaking, Computer Vision is improving the ability to count people and perform pedestrian analysis reliably, efficiently and at scale. However, as an emerging sector, many Computer Vision systems are constrained by a narrow range of object detection capabilities, as well as reliance on specific hardware, deployment requirements and data processing techniques. These limitations – chiefly spawned from the market’s embryonic, vendor-centric efforts – reduce Computer Vision’s potential flexibility, extensibility and ability to generate ROI. This fragmented approach to Computer Vision causes a number of common drawbacks. To find out what those drawbacks are, and how to overcome them, simply download our white paper, People Counting and Pedestrian Analysis with Computer Vision.

You May Also Like…

VisualCortex in the News

VisualCortex in the News

Jul 7, 2022 | ,

VisualCortex’s official launch of its Video Analytics Platform has been covered by a range of trade press outlets in the data, AI and computer vision space. A summation of that coverage, and links to the full write-ups themselves, can be found here:

Loading...