Skip to main content

Detection

Lookout+ uses advanced neural network models to detect and classify objects in maritime environments. The system analyzes video streams in real-time to identify vessels, marine life, navigation markers, and potential hazards.

How Detection Works​

Lookout+ employs deep learning algorithms that have been specifically trained for maritime environments. The detection process involves:

  1. Frame Analysis: Each video frame is processed by neural networks
  2. Bounding Box Generation: Objects are identified and enclosed in rectangular bounding boxes
  3. Classification: Each detected object is assigned a most likely category
  4. Confidence Scoring: Two probability scores are generated for each detection

Neural Network Models​

Lookout+ uses specialized models optimized for different sensor types:

Electro-Optical (Color) Models​

  • Optimized for: Daylight conditions and good visibility
  • Detects: Vessels, marine mammals, navigation markers, swimmers
  • Strengths: High detail recognition, color-based classification
  • Best Performance: Clear weather, adequate lighting
Model Training

Greenroom continuously fine-tunes these models based on user feedback and data you provide. This ongoing improvement process enhances detection accuracy for your specific operational environment.

Detection Confidence​

Each detection includes two critical probability scores:

Existence Probability​

The confidence that an object actually exists within the bounding box.

  • High Probability (>80%): Strong confidence an object is present
  • Medium Probability (50-80%): Moderate confidence, may require verification
  • Low Probability (<50%): Uncertain detection, often filtered out

Classification Certainty​

The confidence that the detected object belongs to a specific category (e.g., ship, whale, buoy).

  • High Certainty (>90%): Strong confidence in object type
  • Medium Certainty (70-90%): Reasonable confidence, may show multiple possibilities
  • Low Certainty (<70%): Uncertain classification, generic labels used

Detection Capabilities by Version​

The range of detectable objects varies by your Lookout+ version:

Basic Detection Categories​

  • Boats: Generic vessel detection
  • Ships: Large vessel identification
  • Marine Mammals: Generic marine life
Assist Focus

Optimized for basic collision avoidance and general situational awareness.

Adjusting Detection Sensitivity​

Under certain conditions, you may need to adjust the detection thresholds:

When to Increase Sensitivity (Lower Thresholds)​

  • High-risk areas: Near shipping lanes or marine protected areas
  • Poor visibility: Fog, rain, or low-light conditions
  • Critical missions: Search and rescue operations
  • Research applications: Marine mammal observation

When to Decrease Sensitivity (Raise Thresholds)​

  • High false alarm rates: Excessive detections from waves, birds, or debris
  • Cluttered environments: Busy harbors with many small objects
  • Performance optimization: Reducing computational load
  • Specific focus areas: When only high-confidence detections are needed
Threshold Adjustment

Changing detection thresholds affects the balance between detection sensitivity and false alarms. Consider your operational requirements carefully before adjusting these settings.

Performance Factors​

Detection accuracy is influenced by several environmental and technical factors:

Optimal Conditions​

  • Clear weather with good visibility
  • Adequate lighting (for EO models)
  • Stable vessel platform
  • Objects within optimal detection range

Challenging Conditions​

  • Poor weather (fog, rain, heavy seas)
  • Extreme lighting (direct sun, complete darkness)
  • High vessel motion or vibration
  • Objects at maximum detection range or near horizon
Best Practices
  • Use both EO and IR models when available for comprehensive coverage
  • Regularly review and adjust confidence thresholds based on operational experience
  • Provide feedback to Greenroom on detection performance to improve model accuracy