AGP Picks
View all

Machine vision is forcing a rethink of video compression and network design

Oct. 29, 2026
Machine vision is forcing a rethink of video compression and network design
By AI, Created 18:09 UTC, Oct 29, 2026, AGP -

Artificial intelligence systems that rely on cameras and other visual sensors are pushing video infrastructure toward new compression and computing models. Industry leaders say machine-oriented approaches could cut bandwidth, shift processing to the edge and change how autonomous and industrial systems are built.

Why it matters: - Physical AI systems need video for machines, not video for human viewers. - That shift could reduce bandwidth, power use and processing overhead across autonomous vehicles, robotics, traffic systems and infrastructure inspection. - The outcome could reshape how networks, chips and edge systems are designed for machine vision.

What happened: - A recent executive vidcast for journalists hosted by the Media Coding Industry Forum featured Aytac Biber of Qualcomm, Honglei Zhang of Nokia and Rob Moore of OP Solutions. - The discussion focused on machine-oriented video compression and how it could support physical AI. - The conversation centered on two emerging approaches: Feature Coding for Machines and Versatile Video Coding for Machines.

The details: - Conventional codecs were built to preserve image quality for human viewers, not to optimize machine inference. - Industrial cameras, autonomous vehicles, drones and robots increasingly use visual data to identify objects, track motion and support automated decisions. - In many machine uses, extra visual detail adds little value and can waste bandwidth and power. - Conventional workflows typically involve camera encoding, transmission, decoding and then neural-network analysis. - That pipeline can create bottlenecks when many cameras feed a centralized system. - Feature Coding for Machines shifts feature extraction to the edge and sends machine-readable features instead of pixel data. - Moore cited comparisons in which FCM matched object detection and tracking accuracy while using 18 to 25 times less bandwidth than traditional codecs. - Zhang said machine vision performance must be measured by more than image quality, including detection accuracy, processing efficiency and the reliability of downstream decisions. - Moore estimated that an autonomous vehicle with about 17 cameras could generate roughly 5.5 terabytes of raw video data per hour. - That figure reflects potential data generation, not necessarily traffic sent across external networks. - Biber said the infrastructure challenge extends to local networks, edge servers, memory, storage and power systems. - Machine-oriented compression could move more processing closer to cameras and sensors. - Biber said machine vision is becoming a growing factor in chip design, including camera processing, AI accelerators, memory, video engines, connectivity and security. - Moore described a future in which intelligent edge devices communicate directly, such as autonomous vehicles sharing road-hazard information without sending full video streams. - Coordinated robots could also exchange relevant observations to better understand an operating environment. - Biber said transmitting extracted features rather than full images could reduce the amount of sensitive visual data leaving a device. - Biber also cautioned that feature-based compression does not guarantee privacy by itself. - Zhang said VVC for Machines keeps a conventional video representation but adds methods to improve machine analysis, including neural-network preprocessing and post-processing. - Zhang said that path may be easier for organizations with existing video infrastructure, but it can preserve some of the overhead of traditional codecs. - FCM compresses intermediate neural-network features for downstream machine-learning systems. - Zhang said FCM can offer greater compression efficiency and more flexibility in distributing compute, but it depends more directly on AI model and application requirements. - Zhang said technical work supporting VVC for Machines is available and FCM standardization has reached an advanced stage.

Between the lines: - The industry is moving from optimizing video for appearance to optimizing it for machine decisions. - That changes the main design trade-off from picture fidelity to inference accuracy, bandwidth efficiency and workload placement. - The choice between FCM and VVC for Machines also shows a familiar split between a more disruptive redesign and a more incremental migration path.

What's next: - The market still needs broader collaboration, implementation feedback and interoperability work. - Future systems will likely be judged on how well they balance accuracy, compatibility, compute cost and network load. - Organizations building physical AI will need to consider compression choices as part of the full system architecture, not as a separate video function. - The long-term test will be whether machine-oriented compression can scale from individual devices to coordinated networks of autonomous machines.

The bottom line: - Machine vision is pushing video technology beyond human viewing standards and toward a new stack built around features, not pixels.

Disclaimer: This article was produced by AGP Wire with the assistance of artificial intelligence based on original source content and has been refined to improve clarity, structure, and readability. This content is provided on an “as is” basis. While care has been taken in its preparation, it may contain inaccuracies or omissions, and readers should consult the original source and independently verify key information where appropriate. This content is for informational purposes only and does not constitute legal, financial, investment, or other professional advice.

Sign up for:

Media Globe Today

The daily local news briefing you can trust. Every day. Subscribe now.

By signing up, you agree to our Terms & Conditions.

Share this page:

Advanced Search Options

Search for:

Search scope:

Type:

Search in:

Date range:

The last

Sort by:

Sign up for:

Media Globe Today

The daily local news briefing you can trust. Every day. Subscribe now.

By signing up, you agree to our Terms & Conditions.