Ovis Edge AI Camera Module Ships 1.5 TOPS for $59

A 28 mm Cube That Sees in the Dark

Edge AI vision hardware has spent a few years stuck between two unappealing options: a Raspberry Pi with a camera bolted on, which works but is bulky and power-hungry, or a sealed commercial module with a closed SDK and no schematic. AIMORELOGY's Ovis lands in the gap. It is a 28 × 28 × 31.7 mm camera module built on the CVITEK CV1842H-P, it does 1.5 TOPS of INT8 inference on-device, and both the hardware design kit and the SDK are published.

• Silicon: CVITEK CV1842H-P with an Arm Cortex-A53 at 1.1 GHz alongside a RISC-V C906 core at 800 MHz
• AI performance: 1.5 TOPS at INT8 with BF16 support, capturing 1920 × 1080 at 60 FPS
• Sensor: SmartSens SC235HAI delivering 1080p full-colour night vision, with an AI-ISP handling the processing
• Price: about $59 for the standard kit and roughly $69 for the CVBS variant, with shipping estimated for December 2026, per CNX Software

What Can 1.5 TOPS Actually Do?

It is worth being precise about this number, because TOPS figures get thrown around carelessly. 1.5 TOPS is not going to run a vision-language model. What it comfortably handles is the classical detection workload: AIMORELOGY lists human and vehicle detection, face detection, pose recognition and object tracking, with YOLOv8n as the reference model. That is the entire feature set most people actually want from a camera, and running it on the module means no video ever has to leave the device.

The dual-core split is the neat part of the design. Linux 5.10 runs on the Cortex-A53 for networking, the web interface and model execution, while RT-Thread runs on the RISC-V C906 for the real-time work. It is a sensible division that shows up increasingly often in this class of RISC-V and Arm hybrid silicon, and it is the reason the module can hold a deterministic capture pipeline while Linux does housekeeping above it.

Memory is the constraint to plan around. 256 MB of DDR3 and 256 MB of SPI NAND flash is generous for an embedded camera and tight for anything ambitious. Choose models accordingly.

Why Does Full-Colour Night Vision Matter Here?

Most low-cost security cameras fall back to infrared after dark, which means monochrome images and a visible IR glow. The SmartSens SC235HAI paired with the AI-ISP keeps colour at night instead, and colour is not a cosmetic detail for a detection pipeline — a model trained on daylight footage degrades badly on monochrome IR frames, and colour is often the feature that separates a parked car from a shadow.

Connectivity is deliberately plain: USB 2.0, 100 Mbps Ethernet, UART and MIPI CSI. No Wi-Fi on board, which will disappoint some people and reassure anyone deploying these on a wired segment. Video encoding covers H.265, H.264 and MJPEG, so it drops into existing NVR setups without translation.

How Open Is It, Really?

More open than most. AIMORELOGY publishes the hardware design resources so expansion boards can be built against the module, and the SDK sits on GitHub with a Docker build environment — which, for anyone who has fought a vendor BSP, is the difference between a weekend project and an abandoned one. Configuration runs through a browser-based interface called OVIS Web rather than a proprietary app.

For makers this slots in below the camera projects we have been covering lately — it is a fraction of the cost of the WLV-01 Raspberry Pi 5 camera with interchangeable sensors and aimed at a completely different job than the dual GMSL2 camera HAT for the Pi 5. Where those are platforms for building a camera, Ovis is a finished sensor node you deploy several of. More single-board computer and edge AI hardware in our mini computer coverage.

The usual crowdfunding caveat applies: December 2026 is an estimate, and estimates on hardware campaigns move. The silicon is shipping in other products and the software is already public, which is a better starting position than most.

Sources: CNX Software — September 16, 2026; AIMORELOGY — September 2026.