System-on-Module

Video Processing Module

A compact System-on-Module that brings 4K video, edge AI and secure, low-latency streaming to any platform.

Built on Ambarella, with full Linux and a simple API, so you can build your application the way you want to.

Request evaluation See the AI in action

Key benefits

Small module, big capability

Compact and light

35 mm square, with standard FPV mounting, to fit where other video systems can't.

Low power

A few watts including the sensor, ideal for battery-powered and unmanned platforms.

4K video

Sharp, efficient H.264 and H.265 video from one or two sensors.

Ultra-low latency

Around 50 ms from sensor to screen, for real-time control.

Any link

Reliable streaming over IP radio, cellular and satellite, even on poor networks.

Secure

Encrypted video, end to end.

Edge AI on Ambarella

Run your AI on the module

Bring models from the popular frameworks. Our conversion tool prepares them for the CVflow AI engine, so detection runs on the module itself, with no cloud and no extra hardware.

PyTorchTensorFlowTF LiteONNXCaffeKeras

Proven models YOLO, SSD, MobileNet, ResNet and segmentation networks.

# Example output: YOLO running on the module Frame 88 · 5 detections (normalised boxes) [56] 'chair', 0.59 [0.454, 0.485, 0.555, 0.548] [39] 'bottle', 0.53 [0.575, 0.401, 0.627, 0.644] [60] 'dining table', 0.44 [0.003, 0.527, 0.402, 0.996] [56] 'chair', 0.29 [0.216, 0.771, 0.260, 0.991] [45] 'bowl', 0.27 [0.000, 0.688, 0.073, 0.808] [YOLO] get_frame: 0.95 ms [YOLO] preprocess: 2.41 ms [YOLO] inference: 85.16 ms [YOLO] postprocess: 15.23 ms

Example detection log from the module. Each line gives the object class, confidence score and bounding box.

Software

Vision-first C/C++ API

API driven, with vision as the first priority.

A simple C/C++ API with minimal dependencies, so you build your application the way you want to.

Software stack

Your application

Endpoint C/C++ APIvideo · image · storage · networking · I/O

Linux on Ambarellaimage DSP · CVflow AI · hardware encoder

Minimal external dependencies

Use case

Object detection at the edge

Detection, tracking, classification and a decision, all in real time and all on the module.

Detect

Find small, distant objects in cluttered scenes.

Track

Follow each object from frame to frame.

Classify

Separate targets from birds and clutter.

Decide

Alert, move a PTZ or hand to an operator.

Use cases

One module, many products

Robotics

Unmanned systems

Security

AI / analytics

Smart cameras / IoT

Appliances

Infrastructure

Automotive

Pair the module with a sensor

A low-light EO sensor, a 640×512 thermal sensor, or both together.

Sensor options Talk to us