The Challenge
A long-running 3D print can fail without immediate notice. I wanted a monitoring setup that could process a live camera feed locally, make print progress easier to check and explore computer-vision-assisted intervention without depending on a cloud service.
My Approach
Camera & Streaming
- Connected a Wyze Cam v4 through Wyze Bridge and MediaMTX, with the camera feed available locally over RTSP.
- Verified H.264 video output at 2560 × 1440 and checked its suitability for print monitoring.
Raspberry Pi & AI Processing
- Used a Raspberry Pi 4 (8 GB) as the local service host, running Docker-based monitoring components.
- Ran and evaluated an ONNX inference pipeline for detecting signs of failed prints from camera frames.
- Connected monitoring services with the printer's Moonraker interface and tested pause/resume control commands.
Validation
- Checked video availability, inference processing and printer communication as separate parts of the workflow.
- Continued testing camera placement, detection reliability and false positives before enabling fully automatic intervention.
Results & Current Status
The project now brings video streaming, local inference and printer-control testing into one engineering environment. These working components provide a foundation for smarter, more practical monitoring of long print jobs.
- Local processing. The monitoring workflow can run on the Raspberry Pi instead of sending inference work to a remote service.
- Clearer observation. The high-resolution camera feed makes the printer easier to inspect remotely.
- Controlled experimentation. Verified pause/resume communication provides a path toward intervention, with safeguards still being tested.
Automatic failure prevention is not presented as a completed feature: detection thresholds, false-positive handling and unattended pausing are still under evaluation.