m1_objectdetection

Dependency for JUCE projects to add image or video object detection for auto-panning use cases after a user has supplied a reference object image for the target object to be detected and tracked.

Implementation

This module provides a JUCE-compatible object detection system using template matching algorithms. The implementation includes:

Core Classes

• Mach1::ObjectDetector: Main detection class
• Mach1::DetectedObject: Structure representing a detected object with position, bounds, and confidence

Key Features

• Template Matching: Uses normalized cross-correlation for object detection
• Proximity-Based Tracking: Weights detections based on proximity to last known position
• JUCE Integration: Native support for juce::Image and JUCE data structures
• Configurable Parameters: Adjustable confidence threshold and proximity weighting
• Multi-Object Detection: Support for detecting multiple instances of the reference object

Usage

#include "m1_objectdetection.h"

// Create detector
Mach1::ObjectDetector detector;

// Set reference object from user selection
juce::Rectangle<int> selection(100, 100, 50, 50);
juce::Image referenceObject = videoFrame.getClippedImage(selection);
detector.setReferenceObject(referenceObject);

// Process video frames
for (auto& frame : videoFrames) {
    auto objectCenter = detector.detectObjectCenter(frame);
    if (!objectCenter.isOrigin()) {
        // Use objectCenter for auto-panning
        float panPosition = (objectCenter.getX() / frame.getWidth() - 0.5f) * 2.0f;
    }
}

Configuration

• setConfidenceThreshold(float): Set minimum confidence for valid detections (0.0-1.0)
• setProximityWeight(float): Set weight for proximity-based tracking (0.0-1.0)
• resetTracking(): Reset tracking state