Detect Objects (Ultralytics Yolo)

Use This When

• Building real-time detection systems where YOLO speed/accuracy tradeoff is optimal
• Deploying custom YOLO models trained with Ultralytics framework
• Creating surveillance or monitoring pipelines requiring fast inference on edge devices
• Standardizing on YOLO architecture for consistent detection across multiple cameras

What It Does

• Loads Ultralytics YOLO model (.pt file) and runs inference on BGR images
• Returns bounding boxes with class labels and confidence scores above threshold
• Supports configurable image size, IOU threshold, and class filtering
• Handles model variants (YOLOv8n/s/m/l/x) with automatic architecture detection

Works Best With

• High FPS camera feeds → this component → track-objects-yolo for multi-object tracking
• Person detections → detect-landmarks for pose estimation and gesture analysis
• Vehicle/object detections → check-object-distance for proximity monitoring
• Integration with segment-image-ultralytics-yolo when both detection and segmentation needed

Caveats

• Model size choice (nano/small/medium/large/xlarge) critically affects speed vs accuracy balance
• Image size parameter directly trades inference speed for detection quality on small objects
• GPU strongly recommended; CPU inference adds significant latency for real-time applications
• Class filtering configured at inference time; changing classes requires model retraining