Alert Mode is an optional feature in Live Video Captioning that enables alert-style visual feedback for binary classification prompts. When enabled, the application provides enhanced visual indicators for "Yes" or "No" responses, making it ideal for automated monitoring and surveillance scenarios.
In Alert Mode, the application:
- Changes the default prompt to: "Is there an accident in the stream? Just Answer with a Yes or No"
- Applies distinct visual styling to "Yes" and "No" responses in the caption display
- Updates the application title to "Live Video Captioning and Alerts"
- Provides clear alert indicators for detections (Green vs Red)
This mode is particularly useful for:
- Accident Detection: Monitor traffic streams for incidents
- Security Monitoring: Detect unauthorized access or suspicious activity
- Safety Compliance: Verify safety protocols are being followed
- Anomaly Detection: Identify unusual events in video feeds
Add or modify the ALERT_MODE variable in your .env file:
# .env file
WHIP_SERVER_IP=mediamtx
WHIP_SERVER_PORT=8889
WHIP_SERVER_TIMEOUT=30s
PROJECT_NAME=live-captioning
HOST_IP=<HOST_IP>
EVAM_HOST_PORT=8040
EVAM_PORT=8080
DASHBOARD_PORT=4173
WEBRTC_PEER_ID=stream
ALERT_MODE=True # Enable Alert ModeWhen Alert Mode is enabled:
| Response | Visual Style |
|---|---|
| Yes | Red/Alert highlighting indicating a positive detection |
| No | Green/Normal highlighting indicating no detection |
While Alert Mode sets a default accident detection prompt, you can customize the prompt in the dashboard UI to suit your specific use case. The key requirement is that your prompt should elicit a "Yes" or "No" response for proper alert styling.
Example prompts for different scenarios:
- Fire Detection: "Is there a fire or smoke visible in the stream? Just Answer with a Yes or No"
- Crowd Detection: "Is there a large crowd gathering? Just Answer with a Yes or No"
- Vehicle Detection: "Is there a stopped vehicle blocking the road? Just Answer with a Yes or No"
- PPE Compliance: "Is the person wearing a safety helmet? Just Answer with a Yes or No"
- Verify the
ALERT_MODEenvironment variable is set correctly in your.envfile - Ensure Docker Compose picks up the environment variable:
docker compose down docker compose up
- Check the application title - it should display "Live Video Captioning and Alerts"
- Ensure your prompt is designed to receive "Yes" or "No" responses
- Check that the VLM model is generating clear binary responses
- Verify the metadata stream is connected (check the status indicator)
- Get Started - Basic setup and configuration
- API Reference - REST API documentation
- System Requirements - Hardware and software requirements