Skip to content

SOD_REALNET.md

Latest commit

 

History

History
105 lines (73 loc) · 3.2 KB

SOD_REALNET.md

File metadata and controls

105 lines (73 loc) · 3.2 KB
Raw

SOD RealNet Integration

This document describes the integration of SOD RealNet face detection into the LightNVR system.

Overview

SOD RealNet is a real-time face detection system that is part of the SOD library. It provides fast and accurate face detection capabilities that can be used for video surveillance and security applications.

Requirements

  • SOD library with RealNet support (directly linked)
  • RealNet model file (e.g., face.realnet.sod)

Integration

The SOD RealNet integration consists of the following components:

  1. SOD RealNet Interface (include/video/sod_realnet.h and src/video/sod_realnet.c): Provides a wrapper around the SOD RealNet API for loading models and performing detection.

  2. Detection System Integration (src/video/detection.c): Integrates SOD RealNet into the existing detection system, allowing it to be used alongside other detection methods.

  3. Test Program (tests/test_sod_realnet.c): A standalone program for testing SOD RealNet face detection on images.

Usage

Loading a RealNet Model

#include "video/detection.h"

// Initialize detection system
init_detection_system();

// Load RealNet model
detection_model_t model = load_detection_model("path/to/face.realnet.sod", 5.0f);
if (model) {
    // Model loaded successfully
    // ...
    
    // Unload model when done
    unload_detection_model(model);
}

// Shutdown detection system
shutdown_detection_system();

Performing Detection

#include "video/detection.h"
#include "video/detection_result.h"

// Assuming you have a frame in grayscale format
unsigned char *frame_data = ...;
int width = 640;
int height = 480;
int channels = 1; // Grayscale

// Create result structure
detection_result_t result;

// Perform detection
if (detect_objects(model, frame_data, width, height, channels, &result) == 0) {
    // Detection successful
    printf("Detected %d faces\n", result.count);
    
    // Process detections
    for (int i = 0; i < result.count; i++) {
        printf("Face %d: x=%.2f, y=%.2f, width=%.2f, height=%.2f, confidence=%.2f\n",
               i+1,
               result.detections[i].x,
               result.detections[i].y,
               result.detections[i].width,
               result.detections[i].height,
               result.detections[i].confidence);
    }
}

Testing

A test program is provided to test SOD RealNet face detection on images:

# Build the test program
cd build
cmake -DBUILD_TESTS=ON ..
make

# Run the test program
./bin/test_sod_realnet path/to/image.jpg path/to/face.realnet.sod output.jpg

The test program will load the specified image, perform face detection using the specified model, and save the result to the specified output file.

Model Format

SOD RealNet models have the .realnet.sod extension. These models are specifically designed for real-time face detection and are optimized for speed and accuracy.

Performance Considerations

  • SOD RealNet is designed for real-time processing and can achieve high frame rates on modern hardware.
  • For optimal performance, use grayscale images for detection (1 channel).
  • The detection threshold can be adjusted to balance between detection accuracy and false positives. A typical value is 5.0.