Lightning Talk: If Zephyr Wants To Power AI Cameras, What Must Change?
Lightningat Open Source Summit India 2026 · on Tuesday, 2026-06-16· at 18:30· Lotus Ballroom 2 (Level 3)·
summary
Zephyr’s video API lacks per-frame metadata and graceful stream shutdown, which causes NPU inference to target the wrong display frame (observed as N+2 frame lag). Prototype introduces a meta buffer pointer carrying ISP statistics and an interference result per frame plus a per-buffer callback that runs in threaded context to avoid CPU wakeups and context switches. Implementation runs as an STM prototype, reduces pointer copies, files an RFC, and is validated but not yet mainline.
AI-written summary · about
chapters (9)
- 0:00Zephyr for AI cameras
- 1:49Zephyr video pipeline limits
- 3:14Metadata and buffer gaps
- 6:03Metadata buffer solution
- 9:34Per-buffer callback proposal
- 12:01Kernel implementation rationale
- 13:26RFC and community discussion
- 14:29Future of AI cameras in Zephyr
- 15:32Q&A
related talks
07mLightning Talk: Zephyr Portability with an AI Application on Very Different MCUs
10mLightning Talk: Debugging Edge AI on Zephyr and Lessons Learned
01h 19mMini-conference: RISC-V Support in Zephyr
14mKeynote: Leveraging Zephyr and ML to Bring Smart Devices to Market, Faster
14mLightning Talk: Every Millisecond Counts: The Fine-tuning Journey of an Ultra-Eff...
42mPorting an AI Powered Wearable Health Monitor to Zephyr on Open Hardware
36mTinyML at the Edge: Deploying and Optimizing AI Workloads on Zephyr RTOS
38mIntroducing a Heterogeneous Core Platform With AI Accelerator for Zephyr Developme...