Onsemi to Improve Detection Time in Vision Systems with their SmartROI feature

Have you noticed how cameras can smoothly capture a specific region within a larger scene? Or how a robot can sort specific objects from a larger bunch on a conveyor belt carrying trash? These tasks are typically accomplished by software running on specific hardware resources within the systems.

onsemi’s Smart Region of Interest (SmartROI) feature within their image sensors allows users & machine vision systems to quickly concentrate on specific areas that need the most attention or analysis. This can lead to optimized industrial processes, more accurate diagnoses and faster decision-making. 



SmartROI in Action with Video Conferencing

Why it’s Important
onsemi’s image sensors with the SmartROI feature can capture more than one region in the scene for every frame. In a single capture process, you can obtain all the needed data – one image at reduced bandwidth and the other specifically targeted for clarity at the original resolution.

In its absence, vision systems must perform the exposure/capture sequence twice. This introduces multiple deficiencies in the vision system:

These deficiencies have serious implications. In a surveillance application, it’s ideal to have the entire scene, albeit at a low resolution to provide complete information of the scene, yet a region focused on the entrance to the door or lock in original resolution and nuanced details. If an operator must achieve this objective through two frames, the latency between the capture of these two frames could render the whole surveillance operation ineffective – could miss the event or the context.

Where to Use SmartROI
Machine vision cameras are typically deployed in factory automation. They operate in a high throughput environment and any delay removed in the operation saves money. By delivering images of more than one region in a single frame, the SmartROI feature enables just that.

Machine vision cameras in general incorporate convolutional neural networks (CNN) and are powered through AI/ML. With both the context of the scene at a reduced bandwidth and detailed image of a pre-determined region available at the same time, cameras with sensors having the SmartROI feature can be trained rapidly and can infer scenes quickly. This reduces critical latencies and increases throughputs and overall efficiency of the vision system.



AR2020 SmartROI – Control Bandwidth & Power While Providing Context

Multiple applications can leverage this feature as listed below.

Factory Automation – Package Scanning
Until recently, package scanning involved scanning just the barcode imprinted or affixed on the boxes. These packages now carry much more than just a barcode. They include product details, company addresses, phone numbers, country of origin, batch and manufacturing information and more. This information may not be needed all at once. SmartROI allows for focusing on just the region of interest.

Factory Automation - Inspection & Quality Control (IQC)
Generally, the objects being scanned have critical regions needing extraordinary focus (e.g., cap on a bottle) while information on other regions (e.g., the branding wrapper that goes around or printed) provides necessary context. With SmartROI, one can get the scan of the entire bottle at a lower resolution while zoning-in on the critical bottle cap for proper closure.

Videoconferencing (VC) – Framing
Today’s VC systems seek a highly immersive experience across all participants. The endpoints, whether they are VC Bars or Tabletop systems with 360-degree views, strive to include every participant, fostering natural and organic face-to-face interactions. SmartROI helps to zone-in on regions where the actions take place, providing a broader view of the person engaged in the action with original detail and the environment of the entire conference.

AR/VR & Gaming
These applications demand a complete view of the surroundings, while the user is focused on the action centered in specific regions. SmartROI helps bring those regions in original detail while providing the rest of the scene in a reduced bandwidth.

Robotics
Vision Guided Articulated Robots are deployed to perform specific tasks. Typically ran by machine learning algorithms, these devices benefit significantly from having the knowledge of the scene in which they operate even as they localize to specific regions related to tasks. In a pick and place operation, the operation could involve focusing on one kind of object amidst several, with the robot’s task being to pick that object type of interest. Once again, SmartROI comes has in handy to help do these tasks very efficiently.

www.onsemi.com