AOV, Black-Light and On-Device VLM: The 2027 In-Vehicle Recorder BOM Is Already on the Show Floor

The organiser's official shortlist quietly published the next-generation in-vehicle bill of materials: black-light AOV vision chips, VLM camera silicon, on-device AI chips, domestic codec SoCs and surveillance-grade storage. The only open question is who can package them for automotive conditions.

Product launches get the photography. The more useful document at Security China 2026 was the organiser's published shortlist of recommended innovative products and solutions. Read it as a component catalogue and the next generation of in-vehicle video hardware is already sitting on the shelf.

The Chip Layer Was Public

The shortlist includes a black-light AOV vision chip series from Goke Microelectronics, a VLM multimodal large-model camera chip and solution plus a domestic public-safety codec SoC from Hisilicon, an on-device AI chip from VSigma, a reconfigurable TPU from Guangzhou Wanxietong, video-surveillance eMMC from Biwin, a 26TB WD Purple Pro surveillance drive, and a fully domestic embedded motherboard from Norco.

It also lists a large-model ultra-low-light checkpoint and traffic camera from Kedacom, large-model products from TVT, Huawei industry-perception super-encoding 2.0 and an AI no-flash smart-traffic solution, an edge analytics terminal, and an intelligent video navigation camera. In other words, sensing, compute, encoding, storage and board-level domestic substitution were all present in one official document.

Industry context, cited here as second-hand reporting that still needs primary verification: a 2025-2026 industry survey described 2025 as the breakout year for AOV low-power always-on recording, with over 60 percent of exhibitors at a major show launching AOV products and a second-generation Smart-AOV design cutting power by half to enable 24/7 recording. Pre-show technical reviews likewise noted large models moving from cloud to device as the core competitiveness of intelligent cameras.

Four Capabilities, Four Old Problems

AOV addresses the blind window after ignition off. Conventional MDVRs depend on vehicle power and go dark when the engine stops — exactly when scrapes, theft and cargo tampering tend to happen. Low-power always-on recording extends driving records into full-time presence, at the cost of redesigning power management, battery protection and storage write endurance. For cold chain, hazardous goods and long-haul freight, that is worth more than another resolution bump.

Black-light imaging and AI-ISP address night-time evidence. Night and tunnels are the high-incident segments for fleets. Ultra-low-light full-colour imaging upgrades seeing into seeing clearly and being usable as evidence; it also links directly to false alarm rates, because image quality sets the input ceiling for any model — a noisy frame will generate false alerts no matter how good the algorithm is.

On-device VLM and edge AI chips address offline availability and scene understanding. Compressing a multimodal model into camera silicon means the unit can make composite judgements locally without shipping full video upstream. For fleets that solves three problems at once: weak-network operation, bandwidth cost, and keeping sensitive driving data inside the vehicle.

Domestic codec SoCs, surveillance-grade storage and local motherboards address cost and supply security. New encoder silicon and modern compression, combined with high-capacity storage and a fully domestic board, set the cost per byte of evidence and determine whether government projects and export programmes can pass localisation or market-access review.

Our Take

Stack those four layers and the competitive axis for in-vehicle MDVR changes entirely: from channel count and megapixels to intelligence per watt, evidence per byte, visibility per night, and trust per alert. These are not independent specs to be stacked. They are four mutually constraining engineering limits — more compute means more power, always-on recording means more storage wear, and pushing ultra-low-light imaging raises noise and false alarms. The vendor that can demonstrate a verifiable balance among them has earned the right to use the word intelligent.

The second signal is that the supply chain is ready. Chips, storage, encoding and motherboards all have domestic options on the official shortlist. That means differentiation no longer comes from being able to buy good silicon; it comes from thermal design, power management, model compression, mechanical reliability and the speed of algorithm iteration — the unglamorous work.

A Seven-Point Selection Checklist

• Ignition-off and parking mode: standby draw, battery protection threshold, continuous recording duration, event wake-up latency.

• Ultra-low light: state the lux condition, whether illumination is active, the colour-to-mono switch point, and the false alarm rate on the night subset.

• On-device capability: number of locally run models and concurrent tasks, first-judgement latency, offline degraded behaviour.

• Encoding and storage: measured bitrate and retention, write endurance and protection, video integrity on sudden power loss.

• Model OTA: grey release, rollback, version traceability, and fallback behaviour when an update fails.

• Automotive and environmental: third-party reports for operating temperature, vibration, supply-voltage fluctuation and EMC.

• Localisation and access: domestic share of chips, board and storage, plus the certification list for target export markets.

Our Judgment

What separates vendors in 2027 will not be who adopted a large model first. It will be who can reliably install these four layers of silicon inside a vehicle that bakes in summer, freezes in winter, suffers voltage swings and constant vibration, and frequently loses its network connection. The technology stack is on a public shelf now; engineering remains the moat.

Sources: the organiser's 2026 innovation products and solutions shortlist (158 companies, 103 products, 87 solutions), pre-show AI model and agent surveys, and second-hand summaries of a 2025-2026 industry survey. Figures on AOV adoption and power reduction are second-hand and should be checked against the primary report before use in procurement. Vendor claims are not independent test results.