Rethinking machine vision in industrial automation

Machine vision has always played a critical role in ensuring safe, efficient, and reliable operation in many industrial settings. However, as vision-enabled machines become more numerous and the type and volume of data they can collect expand, challenges are forcing system makers to look at new approaches to efficiently acquire, process, and utilize visual data.

If we look at the current challenges, they span the spectrum in terms of improving operational efficiency, accuracy, and reliability.

Data overload and processing efficiency that limits throughput are major issues as industries move toward more advanced, faster automation, tasking vision systems with capturing and analyzing vast amounts of data. Traditional vision systems often struggle with the sheer volume of images they capture, much of which can be redundant. The requirement now is not just about capturing high-resolution images but doing so in a way that first and foremost accelerates throughput (in part by minimizing irrelevant data) while maximizing the precision and relevance of the information captured.

Real-time processing is becoming increasingly important, especially in environments where machines need to make instantaneous decisions, such as in quality control or defect detection on production lines. This requires more efficient processing methods and data reduction techniques.

High-speed and high-precision demands increase as production lines get faster. High-speed processing, low latency, and the ability to capture minute changes in a scene in real time are critical. Traditional frame-based systems struggle with motion blur and data overload when capturing fast-moving objects. For example, in applications such as high-speed counting, even the slightest delay in image acquisition and processing can lead to errors.

Sustainability is a growing priority, as many industrial systems operate in environments where power efficiency is key. Vision systems need to operate for extended periods without consuming significant amounts of energy. Traditional image-processing systems, especially those that capture entire frames at a fixed rate, can be power-intensive and require sophisticated cooling or energy management.

Complex lighting and environmental conditions are common in many settings, including extreme brightness, low light, or dynamic lighting scenarios. Vision systems need to cope with high-dynamic-range requirements to capture high-quality images without losing detail in either the darkest or brightest areas. Conventional frame-based systems have struggled in such conditions, leading to the need for more adaptable and sensitive vision technologies.

Predictive maintenance and condition monitoring are growing needs. Vision systems must not only react to issues but also help to predict potential problems before they occur. Predictive maintenance requires vision systems that can monitor machine vibrations, detect wear and tear, and identify early signs of equipment failure.

These challenges point to a more fundamental limitation: Traditional frame-based vision was designed for image capture and human viewing, not for machines that must detect, interpret, and react to changes in real time. As industrial systems move toward higher levels of automation and autonomy, vision is becoming a core component of the perception pipeline.

This shift is driving demand for sensing approaches that reduce latency, limit unnecessary data, and enable faster, more reliable decisions across applications such as monitoring, inspection, counting, and control.

Event-based vision addresses these challenges

Event-based vision, inspired by the human eye and brain, is increasingly used in industrial machine vision to address these challenges. By mimicking biological vision, this technology utilizes efficient sensing and collection techniques that capture changes within a specific scene. This reduces processing requirements compared with traditional frame-based methods while revealing details that conventional systems miss, opening new possibilities for precision and performance in industrial applications.

Event-based vision is particularly suited for industrial automation, IoT, automotive, and edge applications that demand high performance, low power consumption, and operation in challenging lighting conditions. The technology offers significant advantages in speed, power efficiency, dynamic range, and low latency, driving use cases such as high-speed counting, preventive maintenance, and inspection.

From frame-based imaging to event-based perception

In conventional video systems, entire images (i.e., the light intensity at each pixel) are recorded at fixed intervals, known as the frame rate. Standard movies are recorded at 24 fps, with some videos using higher frame rates like 60 fps (16.7-ms intervals). While effective for representing the “real world” on a screen, this method oversamples unchanged parts of an image, especially at high frame rates, while undersampling the most dynamic areas. As a result, critical motion information can be missed between frames.

In contrast, the human eye samples changes up to 1,000× per second without focusing on static backgrounds at such high frequencies. Event-based sensing offers a biologically inspired solution to this under- and oversampling. Unlike traditional cameras, event sensors don’t use a uniform acquisition rate (frame rate) for all pixels. Instead, each pixel defines its sampling points by reacting to changes in the amount of light it detects. Information about contrast changes is encoded in “events”—data packets containing the pixel’s coordinates and the precise time of the event.

Prophesee’s patented event-based sensors, for instance, allow each pixel to activate intelligently based on detected contrast changes. This enables continuous acquisition of essential motion information at the pixel level. The pixels operate asynchronously (unlike traditional CMOS cameras) and at much higher speeds, as they don’t need to wait for a complete frame before reading data.

The advantages of event sensors include high-speed operation (equivalent to 10,000 fps), extremely efficient power consumption (down to the microwatt range), low latency, reduced data processing requirements (10× to 10,000× less than frame-based systems), and high dynamic range (up to 140 dB).

Because only changes are transmitted, event-based data streams are inherently sparse and temporally precise, allowing downstream processing systems—including AI-based processing—to focus on what matters: motion, variation, and anomalies rather than static background information. These attributes make event-based vision systems suited for a wide range of applications and products.

This technology is being commercialized more widely, such as in Prophesee’s Metavision, which has evolved over the past decade to deliver high performance through integrated hardware and software solutions.

Real-time industrial automation with event-based vision

Event-based vision excels in a variety of industrial automation applications. Typical use cases (see Figure 2) range from object tracking and high-speed counting to predictive maintenance and quality control.

Safety: Object tracking

Event-based vision systems excel at tracking moving objects, leveraging their low data rate and sparse information capabilities. This approach allows for precise object tracking with minimal computational resources, eliminating traditional “blind spots” between frame acquisitions. Additionally, event sensors offer native segmentation, focusing solely on movement and disregarding static backgrounds for improved tracking accuracy and efficiency. Event-based vision enhances safety by monitoring worker and machine interactions in real time, even in complex lighting, without capturing images.

Productivity: high-speed counting

Real-time vision systems powered by event-based sensing enable objects to be counted at unprecedented speeds with high accuracy and minimal motion blur. Sensors independently trigger each pixel as objects pass through the field of view, achieving a throughput of over 1,000 objects per second and an accuracy of more than 99.5%, ensuring rapid and precise counting in high-speed environments.

Predictive maintenance: vibration monitoring

Event-based vision enables continuous, remote vibration monitoring with pixel-level precision. By tracking the temporal evolution of each pixel in the scene, the sensors record each event’s coordinates, polarity of change, and exact timestamp. This data provides valuable insights into vibration patterns across frequencies from 1 Hz to the kilohertz range, aiding in predictive maintenance.

Quality: particle/object size monitoring

In high-speed production environments, event-based sensing allows for real-time control, counting, and measurement of particle or object sizes on conveyors or channels. The sensors capture instantaneous quality statistics, ensuring accurate process control at speeds of up to 500,000 pixels per second with a counting precision of 99%, optimizing quality assurance in production lines.

Quality control

Event-based vision systems help lower reject rates with real-time feedback and advanced processing down to a 5-µs time resolution and blur-free asynchronous event output. One specific use case is in the automatic detection and classification of the finest imperfections in manufacturing materials—for example, in automotive parts to perform paint defect inspection, scratch detection, and planarity testing (see Figure 5).

As event-based vision continues to evolve and address diverse market needs, it is establishing itself as a new industry standard. Over the past several years, the technology has expanded to serve a wide array of applications.

Thousands of product developers are now adopting event-based vision for sophisticated camera and perception systems, supported by open-source technology and a growing inventors’ community. These advancements are transforming how machines perceive, process, and react to visual information in real time, bringing greater precision, efficiency, and intelligence to industrial automation operations.

Thibaut Willeman is head of business development and go-to-market at Prophesee, where he works on the market development of event-based vision systems for industrial automation, robotics, and defense applications. He previously held strategy and innovation roles at companies such as Boston Consulting Group, working on growth strategy, product strategy, and innovation initiatives for industrial and technology companies. He holds an engineering degree and a master’s degree in innovation and technology management.

The post Rethinking machine vision in industrial automation appeared first on EDN.