Architecture

Overview

This diagram shows the major parts of the wasmVision system architecture and how they fit together.

flowchart TD
    subgraph wasmVision
        subgraph engine
            subgraph Capture
                Devices
            end
            Runtime[WASM Runtime]
            Capture--frame-->Runtime
            Capture<-->OpenCV
            Runtime<-->Models
            OpenCV<-->Models
            OpenCV<-->CUDA
            subgraph Platform
                logging
                config
                http
                datastore
                time
            end
            Runtime<-->MCP[MCP Server]
        end
        subgraph processors
            Runtime--frame-->processor1.wasm
            Runtime--frame-->processor2.wasm
            Runtime--frame-->processor3.wasm
            Runtime--frame-->processor4.wasm
        end
        processor1.wasm-->logging
        processor2.wasm-->logging
        processor2.wasm-->datastore
    end

Engine

The wasmVision engine. Includes platform capabilities such as MJPEG streaming and saving to files.

Capture

This is how wasmVision can capture or import images or video to be processed from a connected webcam, from files, or from streams. For more information see Capture.

Devices

Specific hardware or software devices that capture images or video. Currently supported devices are a connected camera, a file input, or a stream input using GStreamer.

Runtime

The wasmVision runtime consists of the WebAssembly runtime engine (currently Wazero) and the wasmVision host platform API functions such as logging, configuration, and making http calls to external servers.

Processors

The image processing modules that are used by wasmVision. These are modules written using Go, Rust, or C, and compiled into WebAssembly. They use the wasmCV interface and the wasmVision platform interface. For more information see Processors.

OpenCV

The computer vision processing capabilities implemented using OpenCV with the GoCV language wrappers. The wasmCV programming API is used for this integration.

Models

Machine learning models used for computer vision that can be used directly by processors. There are “well-known” models that can be automatically downloaded by wasmVision when they are needed by a specific Processor. See the Models list for more information.

Platform

The wasmVision platform capabilities provide essential services for processing modules such as logging and access to persistent data. See Platform for more information.

MCP Server

wasmVision includes experimental support for the Model Context Protocol (MCP) by providing a (MCP) Server. See MCP for more information.

CUDA

wasmVision includes experimental support for GPU acceleration using CUDA.

See using wasmVision with CUDA for more information.