Скачать или смотреть Developers Meeting 2025-11-02 — Object Detection Dev Demo

Meeting Summary
Carlos presents object detection updates, we have a discussion on performance, integration, and next steps, and a follow-up demo is given from another developer on the camera monitor app

🧠 1. Object Detection Development (Carlos’s Demo)
A. Overview of Implementation
Uses USB camera (not yet on eCAL interface)
Added multiple detection types (GMM, Vibe, Difference of Gaussian)
Operates through frame differencing of three frames for motion detection

B. Detection Techniques
Gradient, angle, and magnitude calculations for edge detection
Gaussian Mixture Model (GMM): for multi-background filtering
Vibe: faster background learning, fewer “ghosts”
Difference of Gaussian: highlights stars and small bright objects but computationally heavy

⚙️ 2. Technical Challenges and Observations
A. Current Limitations
Low-light performance due to USB camera quality
High computation time for Difference of Gaussian method
Ghost effects when background model updates slowly

B. Optimization Ideas
Adjust Gaussian parameters and thresholds
Separate update threads for non-blocking processing
Possible reimplementation for better OpenCV performance

🔗 3. Integration Plans
A. System Integration
Object detection to be part of camera monitor node
Output: list of detected objects with contours and bounding boxes
Enables object classification and PTF (Pan-Tilt-Focus) camera control

B. Future Expansion
Integration with eCAL framework for real-time data broadcasting
Allow external tuning of parameters (dynamic threshold adjustments)
Sensor fusion and neural network adaptation planned

📸 4. Frame Rate & Exposure Discussion
A. FPS vs Exposure Debate
Question raised about tracking fast-moving objects
Carlos explains limitations of low-FPS USB camera
Long exposure preferred for fisheye (hemisphere) cameras for streak detection

B. Tracking Logic
High FPS needed only for scope cameras
Decision logic based on event lifetime and hardware speed
System avoids tracking objects too fast for PTF movement

🖥️ 5. Hardware & Processing Considerations
A. Performance Management
Distribution of tasks across multiple devices (fisheye, PTF, processing unit)
Plan to use onboard NPUs for machine learning inference

B. Real-Time Processing Pipeline
ECAL nodes handle camera feeds and object detection
Future: machine learning node for classification and adaptive control

👨‍💻 6. Camera Monitor Application Demo (Second Developer)
A. Interface Enhancements
Improved readability and color-coded UI
Added histogram and mask editor
Introduced gamma correction and alternate debayering for infrared cameras

B. Development Tools
Ring buffer for frame-by-frame analysis
Track lines for following moving objects
Layer toggles for ADS-B plane tracking and orientation alignment

🛠️ 7. Mask Editor Features
A. Purpose & Functionality
Ensures privacy by masking private areas (e.g., neighboring buildings)
Reduces computational load and false detections
Editable interface with drawing, erasing, filling, and zooming tools

B. Implementation Details
Left-click: paint mask; right-click: erase
CTRL for fill, SHIFT for straight lines
20-level undo system for precision editing

🚀 8. Next Steps & Coordination
A. Immediate Plans
Carlos to push pull request and merge new object detection code
Integration of eCAL-related features into the camera monitor
Continue collaboration on background subtraction and visualization

B. Future Goals
Add more sensors and heartbeat (GPS time, connectivity)
Implement astronomical calibration (bias and dark frames)
Optimize code performance and expand real-time detection reliability