Learning Mode (Active Monitoring)

Learning Mode, also called Active Monitoring, enables continuous frame capture and temporal analysis. This allows detection of behaviors that unfold over time, like loitering, crowd formation, or process compliance.

What is Learning Mode?

Unlike trigger-based analysis (single frame when event occurs), Learning Mode:

• Captures multiple frames over time
• Analyzes patterns and changes
• Detects behaviors, not just objects

Use Cases

Behavior	Description	Configuration
Loitering	Person in area too long	60s duration, 10s interval
Crowd Formation	People gathering over time	120s duration, 15s interval
Queue Growth	Line getting longer	300s duration, 30s interval
Abandonment	Object left behind	120s duration, 20s interval
Process Compliance	Steps followed correctly	180s duration, 10s interval

Configuration

Profile Settings

Enable Active Monitoring in your profile:

Setting	Description	Range
Enabled	Turn on continuous capture	true/false
Interval (ms)	Time between captures	1000-60000
Max Duration (sec)	Total monitoring window	30-600
Max Images	Total images to capture	5-50
Batch Size	Frames sent together	1-10
Resolution	Image quality	TINY to ULTRA

Example: Loitering Detection

Profile: Loitering Alert
Trigger: AOAS Person (to start monitoring)

Active Monitoring:
  Enabled: true
  Interval: 10000  # Every 10 seconds
  Max Duration: 60  # Watch for 1 minute
  Max Images: 6     # 6 frames total
  Batch Size: 3     # Send 3 at a time
  Resolution: BALANCED

Skill:
  Prompt: "Track this person over multiple frames. Determine if
          they are loitering (remaining in approximately the same
          area) or moving through the scene."
  Objects:
    - Loitering (bool, stateful)

Example: Crowd Monitoring

Profile: Crowd Formation Alert
Trigger: Schedule (every 5 minutes)

Active Monitoring:
  Enabled: true
  Interval: 15000  # Every 15 seconds
  Max Duration: 120 # 2 minute window
  Max Images: 8
  Batch Size: 2
  Resolution: LOW

Skill:
  Prompt: "Monitor crowd density over time. Alert if crowd
          size is growing or exceeds safe capacity."
  Questions:
    - "Current count?" (int)
    - "Trend?" (string: growing/stable/shrinking)
    - "Capacity concern?" (bool)

Trigger Strategies

Schedule Trigger (Continuous)

Poll at regular intervals:

• Use when: Continuous monitoring needed
• Trigger: Schedule
• Interval: Minutes between polls

Example: Queue monitoring every 2 minutes

Event Trigger (On Demand)

Start monitoring when something happens:

• Use when: Resource-efficient monitoring
• Trigger: AOAS Person, Motion, or I/O
• Active Monitoring starts on trigger

Example: Start loitering watch when person detected

Hybrid Approach

Combine both strategies:

1. Schedule polls for baseline monitoring
2. Event triggers for focused attention

Temporal Prompts

Prompt Design for Temporal Analysis

Include temporal context in prompts:

Standard Prompt:

Analyze this image for security concerns.

Temporal Prompt:

You are seeing multiple frames over a 60-second period.
Analyze for:
1. Is the same person present across frames?
2. Are they stationary (loitering) or moving through?
3. Have they made any concerning movements?
4. Compare the first and last frames - what changed?

Frame Referencing

Anava provides temporal context to the AI:

• Frame sequence information
• Time elapsed
• Previous analysis results

The AI can reference:

• "In the first frame..."
• "Compared to earlier..."
• "Over the monitoring period..."

Performance Optimization

Resource Usage

Active Monitoring uses more resources than single-frame analysis:

Setting	Resource Impact
Shorter interval	More frames, higher cost
Longer duration	More data to analyze
Higher resolution	More bandwidth, storage
Larger batch	More memory per analysis

Recommended Configurations

Use Case	Interval	Duration	Images
Loitering	10s	60s	6
Crowd trend	15s	120s	8
Queue length	30s	300s	10
Process step	5s	60s	12

Reducing False Positives

For temporal analysis:

1. Require pattern across multiple frames
2. Use confirmation in prompts ("Must see in 3+ frames")
3. Set appropriate duration for the behavior

VMS Integration

ONVIF Events from Active Monitoring

Events emit when:

• Behavior is confirmed across frames
• Threshold is met (e.g., loitering > 45 seconds)
• Pattern detected (e.g., crowd growing)

Event Timing

Event Type	When Emitted
Detection	When behavior confirmed
Clear	When behavior ends or duration expires
Update	(Optional) As confidence increases

VMS Action Rules

Configure VMS to respond to temporal events:

• Loitering detected → Start high-quality recording
• Crowd alert → Notify operations
• Process violation → Log and alert

Troubleshooting

Monitoring Not Starting

1. Check trigger is firing

   • Verify event reaches ACAP
   • Check profile is active

2. Check configuration

   • Enabled = true
   • Valid interval/duration

3. Check camera connection

   • Device online
   • MQTT connected

Inconsistent Detection

1. Frame quality

   • Increase resolution if details needed
   • Check lighting conditions

2. Interval too long

   • May miss quick behaviors
   • Shorten for rapid changes

3. Duration too short

   • Behavior needs more time to establish
   • Extend for slower patterns

High False Positives

1. Require confirmation

   • Must detect in multiple frames
   • Add to prompt requirements

2. Adjust threshold

   • Higher confidence for events
   • Lower for logging only

Best Practices

Match Duration to Behavior

Behavior	Typical Duration
Quick check	30s
Loitering	60-120s
Crowd forming	120-300s
Process steps	Varies by process

Balance Resources

• Start with longer intervals, shorter durations
• Adjust based on detection needs
• Monitor costs and adjust

Document Configuration

Keep records of:

• Active Monitoring settings
• Detection goals
• False positive rates
• Adjustments made

Related Topics

• Tuning & Optimization - Overall tuning strategies
• Profiles Reference - Profile configuration
• Event Flow - Understanding analysis pipeline