CameraStack on the
Rockchip RK3588.
Ready-made HDMI encoder software for the RK3588: 4K60 HDMI capture, hardware encode, embedded audio, RTSP streaming and a standards-compliant ONVIF interface, with object analytics on the NPU.
The silicon
Inside the
RK3588.
Rockchip's flagship SoC: eight Arm application cores, a neural engine, a Mali GPU, RGA image processing and 8K hardware video codecs, with an HDMI receiver for video and audio input.
- CPU
- 4x Cortex-A76 + 4x Cortex-A55
- up to 2.4 GHz and 1.8 GHz
- AI engine
- 6 TOPS
- three-core RKNN NPU
- HDMI input
- HDMI 2.0 receiver
- up to 4K @ 60 fps, with audio
- Video encode
- H.264 & H.265
- up to 8K @ 30 fps, Rockchip MPP
- Image processing
- Mali-G610 GPU + RGA3
- hardware image processing
On this silicon
What runs today.
- 01 HDMI capture of progressive SDR input at up to 3840x2160 and 60 fps
- 02 Four hardware profiles: a 4K60 main profile and 1080p30, 720p30 and 360p15 secondary profiles, H.264 or H.265 per encoder
- 03 The full RTSP transport family: TCP, UDP, HTTP tunnelling, WebSocket and multicast
- 04 ONVIF Device, DeviceIO, Media, Media2, Events, Analytics and Advanced Security services
- 05 TLS: HTTPS management and snapshots, RTSP over HTTPS, secure WebSocket and rtsps:// streaming, with camera-held certificates managed through the ONVIF TLS Configuration Add-on
- 06 Media Signing: H.264 and H.265 video signed inside the encoded stream with a unique identity for each device, enabled through ONVIF Media2
- 07 Audio: embedded HDMI audio as stereo AAC-LC or mono G.711, configured through ONVIF Media1 and Media2
- 08 Polygon motion detection: ONVIF MotionRegionDetector rules on a reduced analysis image
- 09 NPU object detection with tracked ONVIF scene metadata and selectable models
- 10 JPEG snapshots and an optional 320x240 JPEG stream
- 11 Live encoder control: bitrate and GOP changes mid-stream, on-demand keyframes
- 12 Metadata streaming and events, including the ONVIF ImageTooDark tampering event and SignalLoss
How this port works
An HDMI encoder receives complete pictures from an external source, such as a camera, a computer or a media player. CameraStack captures the RK3588’s HDMI receiver through Linux V4L2, compresses the pictures with Rockchip MPP and serves the streams through the same portable ONVIF and RTSP core as every other platform.
The video path is zero-copy from capture to encode. Pictures stay in DMA buffers throughout, with the Mali GPU and the RGA image processor handling colour conversion and scaling, so the CPU does not touch the video.
Video profiles and input timing
The factory video configuration has four encoders:
| Encoder | Resolution | Frame rate |
|---|---|---|
| Main | 3840x2160 | up to 60 fps |
| Secondary | 1920x1080 | up to 30 fps |
| Secondary | 1280x720 | up to 30 fps |
| Secondary | 640x360 | up to 15 fps |
Each encoder can be set to H.264 Main or H.265 Main through Media2. Profiles can share an encoder, so additional ONVIF profiles do not need another hardware encode. Bitrate and GOP change on the running encoder, and synchronisation-point requests map to on-demand keyframes.
By default each encoder follows the input resolution, up to its configured size, so a 1080p source produces a native 1080p main stream. A client that sets a resolution through ONVIF fixes the output size instead, and that choice is kept when the HDMI input changes.
Supported input covers progressive 16:9 modes from 720p to 3840x2160, plus common PC and SD resolutions, in RGB or YCbCr. When the source changes mode, capture follows the new input and the reported source dimensions update. The ONVIF SignalLoss event reports when the receiver loses the HDMI signal.
Embedded audio and snapshots
The receiver takes stereo LPCM audio from the HDMI source at 32, 44.1 or 48 kHz. CameraStack encodes it as stereo AAC-LC at 48 kHz and 64 or 128 kbps, or downmixes it to 8 kHz mono G.711 mu-law. Audio is exposed through the standard ONVIF audio configurations, which a client attaches to a profile through Media1 or Media2.
Full-resolution JPEG snapshots are taken on demand from the live capture. An optional 320x240 JPEG stream runs at 1 to 5 fps from an RGA3-reduced image, keeping the source aspect ratio with borders where needed.
Motion and object analytics
The Analytics service runs up to four ONVIF MotionRegionDetector polygons over an RGA3-reduced analysis image, evaluated by the same shared detector used on other platforms. The analysis branch also provides the ImageTooDark tampering event. Motion regions follow the input when its resolution changes.
Object detection runs on the RK3588’s NPU through RKNN at five inferences per second. PicoDet-S is the default model, with PP-YOLOE-S and YOLOX-Nano also supported, and YOLOv8n available where its licence permits use and distribution. Detected objects are tracked with stable IDs and published as ONVIF scene metadata.
Built on the Rockchip BSP
The port builds against the Ubuntu Rockchip BSP. Capture runs through the V4L2 HDMI receiver driver, encoding through Rockchip MPP, image processing on the Mali-G610 GPU and RGA3, and inference through the RKNN runtime. For optimal performance the port requires custom patches to the kernel. Everything above the platform plugin, RTSP delivery, the ONVIF services, TLS, media signing, events, tracking and configuration, is the same portable core used on every other platform.
Build on it
Designing an
RK3588 encoder?
The port work is already done, at the level of the Rockchip BSP, MPP and RKNN libraries your board already has. Use CameraStack Server as your complete encoder stack, or as a drop-in ONVIF interface for a pipeline you already have. Evaluation builds are available to teams building real products.
Talk to us about the RK3588