FEATURE Smart factories
HUMAN DETECTION FROM A DIFFERENT PERSPECTIVE
Gabriele Fulco, Product Marketing Manager, Omron Electronic Components Europe, explores what it is that makes humans so difficult to reliably detect, and how successfully navigating these obstacles could usher in a new era of productivity
A
utomatic object detection is growing ever more sophisticated, yet the accurate detection of humans still poses unique challenges. Efforts to mimic human vision to identify objects are nothing new, and recent advances in AI have served to intensify these efforts further. Achieving a computer-based vision system that can not just match but exceed the accuracy and understanding of human vision combined with a human brain has been notoriously difficult. Unlocking this technology could potentially herald a revolution in human progress, revolutionising everything from agriculture to medical science, as well as industrial operations.
Computers have long been able to detect and understand 2D pictures, but dynamic three-dimensional environments are a step far beyond this. Indeed, human vision is not just about simply perceiving the world around us; it is also about understanding it. Our brains are able to constantly provide the vital contextual information to allow us to make sense of our surroundings in real- time. Computers have traditionally been unable to match this level of sophistication, that is until recently. Training a machine to not only perceive but understand the world around it presents complex technological and
20 February 2025 | Automation
computational challenges. The uniqueness and diversity of humans themselves make them one of the most challenging subjects to reliably detect without training any system extensively on specific individuals.
Even a change of clothing or hairstyle can
present problems. When you add in additional factors such as the wider environment with which humans are interacting, combined with the unpredictability of human behaviour, the technical challenges quickly mount up. Any viable solution also has to be cost-effective and economical in size in order to be practical in everyday environments. Solving these problems is not easy. In fast-moving industrial settings for instance, several humans may all be working at speed, carrying out various different duties within the same space. Attempting to track their movement from a side-on or even an isometric view has traditionally proven an imperfect solution, as this requires the system to have an understanding of the depth of vision. In a single-camera configuration, one person can also very easily obscure another from view and create blind spots.
In addition, one of the major challenges in the development of vision sensing technologies is not so much in the capture of images, but in processing them. For a machine to understand human movement in real-time requires a large amount of computational power to ensure high
speed and accuracy. Since no two environments are the same, developing a system that can not only understand the nuances of human movement, but also adapt to different scenes and lighting levels, has traditionally been a barrier to such technologies becoming viable on a wide scale. Omron’s AM1 human detection system tackles these challenges quite literally from a different perspective. Designed for optimising human productivity in industrial settings, it utilises a single top-down camera, combined with software optimised specifically to detect and interpret human movement. In doing so, it can provide a more accurate picture of where in a given space human workers are located, while also reducing the likelihood of overlapping and blind spots. The AM1 software has been trained to understand typical patterns of human movement, and can track up to 10 individuals within a 7m x 7m area with an accuracy exceeding 95 percent.
Having this capability allows organisations to track where and how workers are moving, or how long they’re staying at a particular station for. This information can in turn be used to detect bottlenecks, and ensure that space utilisation and workflows are as efficient as possible. In practice this could mean removing obstacles, or shortening routes that are most frequently used.
AM1’s accuracy is achieved through the system’s 10fps frame rate. Image data from the camera (or multiple cameras) is fed into a processing hub via Ethernet, which is powered by an Intel OpenVINO accelerator. This is allows the system to turn raw data into useful information quickly. Once processed, the information is then conveyed for operators to a standard PC or PLC. Omron’s vast library of data means that the system does not need to be trained on particular individuals, and can detect any human body type. As such, no specific programming skills are required for users. Aside from optimising productivity, other potential uses could involve occupancy detection to determine the appropriate HVAC conditions, or intrusion detection during non- work hours.
Omron Device & Module Solutions Europe
https://components.omron.com/eu-en
automationmagazine.co.uk
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40
