My "Motion detect" doesn't work...

[DevlTz]

I was trying to do the detection only by movement so as not to eat too much space, but as I understand, mine doesn't work... I did the installation with Docker and I don't know if some configuration was missing on my part or if it's a limitation of OS:alma 9.7

I tried to update the system, I tried to change the detection type, I tried to lower the security sensitivity, rtsp... everything, but it didn't work out, does anyone know why?

[DevlTz]

Another thing I would try would be to actually use "Zones Detection", but I can't even visualize the space to be able to draw or choose the scan zones.

[matteius]

ffmpeg needs to be available on CLI for the go2rtc snapshot end-point to be able to retrieve a snapshot.

[DevlTz]

Docker Feedback: FFmpeg missing, Critical crash with stream names, and Motion model issues

Hi! First of all, thanks for reply me. I've spent some time setting up LightNVR via Docker (Compose) and I've some kind of problems.

However, I encountered three major issues that might affect other Docker users. Here is my report and the workarounds I used:

1. Snapshot Failure (Missing FFmpeg dependency)
   The go2rtc snapshot endpoint was failing consistently, which prevented the AI detection from receiving frames.

Error in logs: [ERROR] Failed to fetch snapshot from go2rtc (HTTP 500, size: 51)

Cause: The default Docker image seems to lack ffmpeg. go2rtc requires it available in the CLI path to generate snapshots.

Workaround: I had to manually install it inside the container (apt-get update && apt-get install -y ffmpeg).

2. Critical Crash Loop with Stream Names (Sanitization)
   I accidentally named a stream with parentheses (e.g., Camera (Entrance)). This caused the container to enter a crash loop immediately upon startup.

Error in logs: sh: 1: Syntax error: "(" unexpected

Observation: It seems the stream name is being passed directly to a shell command (likely the HLS or detection script) without sanitization/escaping.

Fix: I had to wipe the lightnvr.db database to recover the system and rename streams using only alphanumeric characters.

3. Native Motion Model Missing
   When detection_model defaults to motion (or is manually selected), the system fails to load the internal model.

Error in logs: [ERROR] MODEL FILE NOT FOUND: motion

Context: The mapped /models directory was initially empty.

My Setup:

OS: AlmaLinux (Host)

Deployment: Docker Compose

Backend: TFLite (via light-object-detect API)

Hope this helps improve the stability for the next release!

[INFO] [saida] Running detection (interval=10s, elapsed=11s, model=motion)
[ERROR] MODEL FILE NOT FOUND: motion
[WARN] [saida] Failed to load detection model: motion

[ERROR] Failed to fetch snapshot from go2rtc (HTTP 500, size: 51)
[WARN] API Detection (snapshot): Failed to get snapshot from go2rtc for stream saida

[INFO] Attempting to terminate previous instance (PID: 20) with SIGTERM
[SIGNAL] Received shutdown signal, initiating shutdown...
sh: 1: Syntax error: "(" unexpected

[DevlTz]

[Bug/Migration] : "table detections has no column named recording_id" even after update

Hi! I ran into an issue after updating my LightNVR Docker container.

The Problem: My recordings were failing to link to detections, and playback in the web UI was broken. The logs showed repeated errors: [ERROR] Failed to prepare statement: table detections has no column named recording_id.

However, the startup logs claimed the database was fine: [INFO] Database schema is up to date (version 16)

The Fix: I had to manually access the database via sqlite3 and run: ALTER TABLE detections ADD COLUMN recording_id INTEGER;

Note: I am using a persistent volume for /var/lib/lightnvr/data. It seems the migration logic didn't trigger for the existing database file, or the version check (v16) returned true even though the column was missing. Just letting you know in case others face this!

• Anyway, NOW I can record with detection!!!!

[matteius]

I had forgotten to include a recent migration, but it did get included in 0.19.4 so I think we should be golden now on that particular issue -- thanks for helping to debug it!