docs/PROJECT_STRUCTURE.md

Project Structure Guide

Overview

NETRA is now organized into clear, logical sections for easy maintainability and scalability.

NETRA/
├── src/                     Source code
│   ├── detectors/          AI detection modules
│   ├── pipeline/           Video processing pipeline
│   └── utils/              Utility functions
├── models/                 Trained AI model weights
├── config/                 Configuration files
├── webapp/                 Flask web application
├── docs/                   Documentation
├── tests/                  Unit tests
└── README.md              Main documentation

Directory Details

/src - Source Code

Contains all core application code organized by functionality.

/src/detectors

All AI detection models wrapped in consistent interfaces:

• yolo_detector.py - Object detection (people, cars, animals)

  • Uses YOLOv8 from Ultralytics
  • Model: models/object_detection/yolov8n.pt

• violence_detector.py - Violence/fight detection

  • Uses Keras CNN trained on violence data
  • Model: models/violence_detection/violence_model.h5

• weapon_detector.py - Gun/knife detection

  • Uses custom trained YOLO model
  • Models: models/weapon_detection/best.pt, models/object_detection/yolov8n.pt

• pose_detector.py - Human pose detection

  • Uses YOLOv11 Pose
  • Model: models/pose_detection/yolo11n-pose.pt

• anomaly_detector.py - Anomalous behavior detection

  • Uses PyTorch anomaly model
  • Model: models/anomaly_detection/best.bin

/src/pipeline

Video input and processing modules:

• video_capture.py - Camera/RTSP stream capture
  • Motion detection using MOG2
  • ROI extraction for efficient processing

/src/utils

Helper utilities and common functions (for future expansion)

/models - Trained Model Weights

Organized by detection type:

models/
├── object_detection/      YOLO object models
├── violence_detection/    Violence detection models
├── weapon_detection/      Weapon detection models
├── pose_detection/        Pose detection models
└── anomaly_detection/     Anomaly detection models

Each folder contains the trained weights ready for inference.

/config - Configuration Files

Centralized configuration management:

• settings.py - Application-wide settings

  • Flask configuration
  • Database settings
  • File upload parameters
  • Session management

• model_config.py - Model-specific configuration

  • All model paths (easy updates)
  • Detection thresholds
  • Processing parameters

/webapp - Flask Web Application

The complete web interface:

webapp/
├── app.py                 Main Flask application
├── templates/             HTML templates
├── static/               CSS, JavaScript, images
├── uploads/              User uploaded videos
├── processed/            Processed video outputs
└── instance/             SQLite database

/docs - Documentation

• ARCHITECTURE.md - System architecture overview
• ANOMALY_DETECTION_GUIDE.md - Anomaly detection setup
• SETUP.md - Installation and setup instructions
• API.md - API reference

/tests - Unit Tests

Test files for core modules (optional for future):

• test_detectors.py - Tests for detector modules
• test_pipeline.py - Tests for video pipeline

Import Examples

Old Structure (Before)

from NETRA.violence_detector import ViolenceDetector
from NETRA.yolo_detector import YOLODetector
from pose_detection import PoseDetection

New Structure (After)

from src.detectors import ViolenceDetector, YOLODetector, PoseDetection
from src.pipeline import VideoCapture
from config import MODEL_PATHS, DETECTION_THRESHOLDS

File Organization Benefits

✅ Clear Separation - Code organized by functionality ✅ Easy Navigation - Developers know where to find things ✅ Scalability - Easy to add new detectors or features ✅ Maintainability - Configuration centralized ✅ Model Management - All models in one organized place ✅ Documentation - All docs in dedicated folder

Configuration Usage

Accessing Model Paths

from config import MODEL_PATHS, get_model_path

# Get specific model path
violence_model = get_model_path('violence')

# Get all available models
from config import get_all_available_models
models = get_all_available_models()
# Output: {'violence': True, 'yolo': True, 'weapon': True, ...}

Accessing Thresholds

from config import DETECTION_THRESHOLDS

yolo_threshold = DETECTION_THRESHOLDS['yolo']  # 0.25
violence_threshold = DETECTION_THRESHOLDS['violence']  # 0.6

Accessing Settings

from config import SECRET_KEY, MAX_CONTENT_LENGTH, UPLOAD_FOLDER

print(f"Max upload: {MAX_CONTENT_LENGTH}")
print(f"Upload location: {UPLOAD_FOLDER}")

Next Steps

1. ✅ New folder structure created
2. ✅ Files reorganized
3. ✅ Configuration centralized
4. 🔄 Update imports in webapp/app.py
5. 🔄 Test all functionality

Troubleshooting

"Module not found" errors:

• Ensure sys.path includes project root
• Check that __init__.py files exist in all packages

Model not loading:

• Check config/model_config.py paths
• Verify model files exist in models/ directory
• Check file permissions

Import issues:

• Run from project root directory
• Ensure Python path includes src/