Everything you need to know about vision systems in 2023

Embedded vision systems are evolving, and it’s worth noting how these changes will likely influence business-critical applications across industries. From agricultural land to retail spaces, the quantum leaps taken by vision systems promise to reshape how machines perceive and interact with environments.

First, let’s look at some camera types that have helped transform how several industries use embedded vision systems to see, analyse, and act.

Types of cameras used in modernised vision systems

3D cameras

3D cameras play a major role in empowering embedded applications with vision power, particularly when depth perception and accurate distance measurement are critical. These include ToF (Time-of-Flight) cameras and AI-based stereo vision cameras. For instance, with the ability to send a light signal and measure the time it takes to return after bouncing off an object, 3D ToF cameras provide precise depth mapping that can be invaluable in autonomous vehicles, robotics, and security systems. Also, the spatial data provided by 3D ToF cameras can be crucial. They help avoid obstacles and optimise path planning, ensuring that robots or vehicles navigate environments.

Stereo vision cameras aim to replicate human vision’s depth perception through triangulation, considering factors like camera distance, resolution, and focal length. By capturing two distinct 2D images from varied positions, these cameras correlate the data to generate a depth image. It makes them well-suited for outdoor applications with expansive fields of view.

Smart AI cameras

Smart AI cameras with AI capabilities empower embedded vision applications to analyse, interpret, and respond to visual inputs on the device itself. This localised processing capability, called edge computing, minimises the need to transmit data to a central server, thereby reducing latency and bandwidth usage. In embedded vision applications such as facial recognition systems, automated quality inspection in manufacturing, or adaptive traffic management systems, smart AI cameras can deliver real-time insights and actions.

Moreover, combining AI and embedded vision allows these applications to learn and adapt to evolving scenarios and requirements. For example, in an industrial setting, smart AI cameras can detect anomalies or defects in products on a production line, refining their detection algorithms to enhance accuracy and reduce false positives.

In retail, these cameras can analyse customer behaviours and preferences, adjusting in-store advertising and layouts to maximise customer engagement and sales, showcasing their adaptability and intelligence.

GMSL2 (Gigabit Multimedia Serial Link 2) cameras

GMSL2 cameras can transmit high-resolution video with low latency over long cable lengths. In scenarios that require real-time video transmission over significant distances, such as autonomous vehicles, remote surveillance systems, or large-scale industrial monitoring systems, GMSL2 technology can be pivotal in ensuring reliable, high-quality video feeds while maintaining a minimal lag, which is crucial for real-time decision-making.

Consider an autonomous vehicle that relies on real-time video feeds from multiple cameras to make split-second decisions. Integrating GMSL2 cameras can enable the vehicle’s embedded system to seamlessly manage and analyse high-quality video inputs from various angles, promoting enhanced environmental awareness and safer decision-making.

So, how do they compare to Ethernet cameras?

The fact is that both GMSL2 and Ethernet cameras address the increasing demands for fast data rates, high bandwidth, data integrity, and improved EMI/EMC performance, each catering to specific applications. GMSL2 cameras, being more advanced, excel in embedded vision applications requiring exceptional speed and performance. Factors to weigh include transmission distance, data transfer speed, EMI/EMC performance, and cost when choosing between them.

HDR cameras

HDR cameras are designed to capture high-quality images, no matter the lighting conditions. An HDR image is obtained by capturing three images of the same scene, each at different shutter speeds. The sensor then combines all the photos to stitch together the entire image. HDR cameras use two modes of image capture, namely:

Multi-exposure or traditional HDR uses short and long exposures to capture various brightness levels, from the brightest highlights to the darkest shadows, in a single image. These are processed and merged to create a final HDR image.

Split-pixel HDR involves splitting each pixel on the sensor into two sub-pixels to capture the brighter and darker areas. Since images are simultaneously captured – not sequentially, this reduces motion blur and delivers a more accurate representation of colours.

For example, HDR cameras power large vehicles that rely on multi-camera systems to generate a 360-degree view of their surroundings. HDR is critical, as multiple cameras must provide clear images in the face of on-road lighting conditions.

Automotive cameras

Automotive cameras are invaluable for applications like ADAS, Autonomous Mobile Robots, and agricultural vehicles, where performance in diverse lighting conditions is critical. Unlike standard cameras, automotive cameras excel in reliability across extreme temperatures, high vibrations, and challenging environments.

These specialised cameras adhere to rigorous industry standards and regulations, undergoing thorough testing to ensure their reliability. They are also engineered for seamless integration into automotive platforms and systems, supporting widely recognised interfaces like GMSL, FPD-Link, or MIPI. This compatibility simplifies their incorporation into existing hardware and software setups.

For instance, automotive cameras in ADAS can detect lane boundaries, traffic signs, pedestrians, and other vehicles. They track the vehicle’s position within the lane, monitor blind spots, and provide collision warnings. The data collected by these cameras enables features like lane-keeping assistance, automatic emergency braking, etc.

Latest use cases of embedded vision systems

Besides those mentioned above, let’s explore other popular embedded vision systems that such cameras have modernised.

Automated weeders

Embedded cameras in automated weeders facilitate precise weed identification and location tracking, enabling targeted herbicide application or physical removal. Integrating intelligent vision systems allows these agricultural robots to differentiate between crops and weeds so that crops are not inadvertently damaged during weeding.

Goods-to-person robots

Embedded camera systems in goods-to-person robots empower them to navigate warehouses and locate items. These robots can identify, fetch, and transport goods through computer vision and machine learning, reducing manual labour and expediting order fulfilment in logistics operations.

Autonomous patrol robots

Embedded cameras enable patrol robots to autonomously monitor and analyse environments for anomalies or security threats. Utilising real-time video analysis, the vision system-based robots can identify potential issues and address them directly or alert human operators.

Teleconferencing systems

Teleconferencing systems utilise embedded cameras to capture video, facilitating virtual communication across distances. These cameras can enhance user experience by providing high-quality video when integrated with smart focusing and tracking features. So, participants are visible and engaged, thus bridging geographical gaps in communication.

ANPR (Automatic Number Plate Recognition) systems

Embedded cameras in ANPR systems extract and analyse license plate information for various applications like traffic enforcement and vehicle tracking. These vision systems can accurately read license plates using optical character recognition, providing crucial data for parking management, security monitoring, etc.

Point-of-Care devices

In point-of-care medical devices, vision systems can be instrumental in diagnostic procedures. They capture images for analysis, thereby enabling telemedicine consultations. Integrating high-quality cameras helps capture images and perform data analysis, providing reliable information to make informed patient care decisions.

Photomicrography devices

In photomicrography devices, embedded cameras capture magnified images seen through a microscope, enabling detailed examination and analysis of microscopic entities. High-resolution cameras help document and share intricate microscopic details, enabling research and analysis in biology, materials science, etc.

Smart cart applications

Embedded vision systems in smart carts enhance the shopping experience by automating item recognition and checkout processes. These systems can identify and tally goods through visual recognition, offering seamless, cashier-less checkout experiences.