AI-Portable
Article image for CVITEK CV1842H-P edge AI camera module offers 1.5 TOPS, 1080p night vision, and $59 Kickstarter price Articles
Not Applicable

CVITEK CV1842H-P edge AI camera module offers 1.5 TOPS, 1080p night vision, and $59 Kickstarter price

The AIMORELOGY Ovis stacks CVITEK's mixed-core CV1842H-P SoC with a SmartSens sensor and optional CVBS output, delivering full-color 1080p night vision and 1.5 TOPS of edge inference in a drone-scale module.

Condensed by AI-Portable from Editorial queue.

A three-board stack built for drone-scale vision

As CNX Software details, the AIMORELOGY Ovis puts the CVITEK CV1842H-P SoC, a SmartSens image sensor, and optional analog video output into a modular sandwich. The Core board measures just 20 × 20 × 5.7 mm and carries the SoC, 2 Gbit (256 MB) SPI NAND flash, 256 MB DDR3 stacked in-package, USB 2.0 through a 4-pin SH0.8 connector, and UART debug pads. A Sensor board adds the SmartSens SC235HAI CMOS sensor, integrated lens mount, 10/100 Mbps Ethernet, and UART, bringing the stack to 28 × 28 × 31.7 mm with lens. An optional CVBS output board uses the MacroSilicon MS7024 encoder for analog composite video. All boards connect through 2× 50-pin board-to-board connectors, so the same Core can be paired with different imaging front ends.

What the CV1842H-P brings to the table

The SoC itself is a mixed-core design: one Arm Cortex-A53 at up to 1.1 GHz runs Linux, while a T-Head C906 RISC-V core at up to 800 MHz handles lighter control tasks under RT-Thread. AI workloads get a 1.5 TOPS INT8 TPU with BF16 support. The AI-ISP can deliver full-color 1080p at 30 fps night vision using AI-based noise reduction (AI-BNR), and the video engine encodes H.265, H.264, or MJPEG at up to 8 MP at 25 fps. In practice, the module can stream as a UVC webcam or over RTSP via USB NCM virtual Ethernet—but not both at once. The browser-based OVIS Web dashboard uses WebUSB to tune ISP, OSD, and inference settings without code.

Software and practical considerations

Ovis boots Linux 5.10 on the Cortex-A53 and RT-Thread on the RISC-V core from the same 256 MB SPI NAND. Built-in AI functions cover:

  • Human and vehicle detection
  • Face detection
  • Pose recognition
  • Single-object tracking

Custom models compile to SOPHGO/CVITEK BModel files, with YOLOv8n documented as a reference. An SDK on GitHub includes U-Boot 2021.10, cvi_mpi, image operators, libsophon, and a Docker build environment. One appealing detail: the CV1842H-P uses the same 10×10 mm BGA pinout as CV1812H-class chips, so existing CV1812H custom boards can swap the SoC without a footprint change. That does not apply to Milk-V Duo boards, which use different chips.

Crowdfunding starts on Kickstarter. The Standard Kit with Core, Sensor Board, and USB cable is HK$460 (about $59), while the CVBS Kit adds analog output for around $69. Shipping is expected in December 2026.

Original source ↗