This blog post was originally published at V-Nova’s website. It is reprinted here with the permission of V-Nova.
In Industrial and Defense environments, I hear the same story. More cameras. Higher resolutions. Stricter latency targets. Infrastructure that cannot be replaced easily. And increasing pressure around storage, bandwidth, compute, and privacy.
This is why MPEG-5 LCEVC is becoming even more relevant. It improves compression efficiency and visual clarity by adding a lightweight enhancement layer containing residual information on top of a base layer. The enhancement layer is then combined with the base layer to reconstruct the full picture. This works with popular codecs such as H.264, HEVC, VVC, and AV1.
But the more interesting shift is that LCEVC is a scalable, hierarchical data format that – based on the type of compute task that must be performed – allows to selectively access, transfer and decode just a lower-resolution portion of the full-resolution video. We call formats like LCEVC “Compute-Aware,” meaning that information is organized for efficient access: rather than being static and flat files, compute-aware data formats are structured and partially queryable. The base layer of an LCEVC-enhanced video is a fully decodable, lower-resolution video on its own. It can be selectively fetched, it’s fast to decode, and it’s ideal for many types of continuous AI inference. The enhancement layer contains the residuals needed to reconstruct the full-fidelity frame when required, ideal for precise recognition, specific training tasks and storage.
Ensure the best accuracy, but let AI decode only what’s needed
This is already validated. In joint tests with Intel on UHD traffic clips,usingIntel’s Deep Learning Streamer and MobileNet SSD model, running inference on the decoded base layer instead of the full resolution stream cut total decode and analytics time by 30 to 50% on CPU. On Intel integrated graphics, inference throughput increased by over 3x,power usage was reduced by 70%,and memory bandwidth by roughly 80%. Crucially, object detection accuracy remained unchanged, since in practice the AI workflow would have downsampled the picture anyway after fetching and decoding the full–resolution video frame. For AI tasks where full-resolution detail is key, LCEVC still providesfaster throughput, thanks to reduced I/O from more efficient compression.
LCEVC is about structuring video data so the pipeline does less unnecessary work. By separating base video from enhancement data, systems can avoid full reprocessing when it isn’t required, reducing compute, power, and cost across the pipeline.
What this unlocks for industrial systems
Always-on inference at lower cost
The decoded base layer is consistent, clean, and predictable. Continuous detection, tracking, and triage can run directly on it. The fully reconstructed picture, created by combining the base with the enhancement-layer residuals, is always available “on demand” for operators or downstream systems that need finer detail.
More streams per node
Less decoding and scaling means more streams for the same CPU, GPU, or edge SoC budget. On Intel integrated graphics, the hardware supported approximately three times more UHD streams in real time when inference operated on the base layer.
More stable throughput, lower latency, and energy consumption
When many cameras compete for compute, hierarchical formats help keep timing predictable. Per-frame processing time fell by about two-thirds in the Intel pipeline, giving analytics systems more headroom for decision logic and alerts. Power consumption per channel also dropped by about 70%, contributing to more sustainable, high-density edge deployments.
Cleaner inputs improve model behavior.
LCEVC delivers higher visual quality at a given bitrate. Even when the model works at base-layer resolution, it receives a cleaner source image. Detection performance stayed essentially flat compared to the non-LCEVC reference, while bitrate and compute dropped significantly.
Where this matters most
Automotive teleoperations and car-to-cloud analytics
Today: H.264 on fixed encoders, strict latency, and unstable cellular conditions.
With LCEVC: Lower bitrate, more stable operator frame rate, and AI triage on the base layer, reducing the bandwidth required for AI processing while preserving decision accuracy.
Perimeter security and counter UAS
Today: Multiple EO (visible-light) and IR (infrared) feeds over limited infrastructure.
With LCEVC: Higher stream density and consistent tracking performance, with full-fidelity reconstruction available as needed.
Defense ISR and control rooms
Today: Narrow RF links carrying multiple sensor feeds.
With LCEVC: More feeds per link and clearer imagery at equivalent bitrate, while also reducing compute per stream.
Privacy is built in rather than bolted on
Industrial video contains people, plates, screens, and sensitive environments. LCEVC’s layered structure helps teams apply privacy rules earlier in the workflow.
The decoded base layer naturally contains less fine detail. The full reconstruction requires the enhancement layer containing residual information, which means access can be gated and audited.
Practical benefits of LCEVC
Minimum necessary sharing
Uses the base layer for broad distribution or first-level review. Reserves full reconstruction for authorized users and finer detail analysis.
Region-aware and time-aware workflows
Carries metadata for sensitive areas or time windows and applies rules without forcing aggressive blurring across the image.
Reduced redaction workload
Investigations and disclosures are accelerated because reviewers can start with the lower-detail base layer.
Examples in the field
Fleet evidence and public safety
Base layer for routine sharing. Full reconstruction for authorized forensics.
Defense trials in mixed environments
Respect privacy constraints while preserving mission-grade clarity where allowed.
How to start
You do not need to be a video specialist. We work side by side with teams in Automotive, Defense, Energy, and Industrial environments to build pipelines that process media files faster, scale to larger volumes, and stay within practical operational constraints.
To explore how LCEVC can fit into your architecture and deliver benefits rapidly, we can walk you through it.
Gianmarco Tasca
GM Industrial Products
V-Nova
