CameraStack on the
HiSilicon Hi3519V101.

How this port works

The Hi3519V101’s camera and encoders are reached through HiSilicon’s MPP SDK. CameraStack handles platforms like this with a C producer: a compact native adapter, written in C straight against the version-matched MPP and sensor libraries, that owns the vendor media pipeline from sensor to hardware encoder and hands encoded video to the portable core. The Go server is statically built for the board’s Armv7 uClibc environment and does not link the vendor SDK; the whole stack runs inside the platform’s 64 MB Linux memory budget, alongside the 448 MB media-memory zone the vendor pipeline owns.

Control flows the other way too. Because the producer owns the encoder, settings changed through ONVIF are applied live: frame rate, bitrate and GOP change mid-stream, a main-profile codec change between H.264 and H.265 rebuilds only the selected VENC channel while the sensor, ISP, sibling stream and snapshot paths keep running, and synchronisation-point requests map to on-demand IDR keyframes. Encoding is demand-driven per profile: each hardware encoder runs only while clients consume its stream, a resumed stream starts on a fresh keyframe, and the resident pipeline keeps snapshots and motion events flowing with both video encoders idle. The proven profiles are 4K at up to 30 fps, H.264 or H.265 Main, and a 1080p H.264 sub-profile, over unicast UDP, interleaved TCP, and wired multicast with AutoStart.

Analytics, Imaging and OSD

The Analytics service runs ONVIF MotionRegionDetector rules: up to four named polygons, each with its own sensitivity and damping. IVE, the chip’s vision engine, computes 4x4 block differences over a resident 320x180 luma branch at 5 samples per second, producing an 80x45 activity grid, and CameraStack rasterises arbitrary ONVIF polygons onto it, so motion detection never scans full video frames on the CPU and keeps running while both video encoders are idle. The motion classifier adapts its threshold to each grid’s noise floor, so detection stays dependable when high sensor gain makes a dark scene noisy. This silicon predates the general-purpose NPUs of CameraStack’s newer ports, so the Analytics service offers hardware-backed motion and tampering events rather than object detection.

ImageTooDark is derived from the live ISP exposure statistics, normalised by exposure time and the analogue, digital and ISP gain terms, so a covered lens stays detected even when automatic exposure raises the gain to compensate. The event holds under both automatic and manual exposure.

The Imaging service operates the running ISP: brightness, contrast, colour saturation and sharpness, automatic or manual exposure with manual exposure time and gain, automatic or manual white balance with independent red and blue gains, and backlight compensation with an adjustable level. Settings persist across a restart.

JPEG snapshots come from persistent hardware encoder channels at native sensor resolution, including while the video encoders are paused. Hardware OSD provides four text slots per profile with live date and time text, standard and custom positions and the full transparency range, and the same OSD state is shared with the corresponding snapshot path, so snapshots and streams always match.

Built on the HiSilicon MPP SDK

The producer builds against HiSilicon’s MPP middleware and the matching sensor library. Capture runs through VI, the video input module that owns the sensor and ISP; VPSS, the video processing subsystem, scales each stream profile, with two live channels feeding VENC 0 and VENC 1, the hardware H.264 and H.265 encoders, extended channels feeding the persistent JPEG encoders, and a low-rate luma channel feeding IVE motion analysis. The producer builds on the same shared common core as CameraStack’s other custom-SDK platforms, so framing, configuration, motion-grid mapping, event formatting, OSD state and pause state are well-tested portable code, with only small Hi3519-specific customisations on top.