⚙️ Product Counting Flow

The Product Counting Flow is a specialized system designed for real-time detection and counting of products moving along conveyor belts or production lines.
Built upon the Rosepetal AI multicamera framework, this flow combines computer vision, intelligent counting logic, and structured session tracking — enabling reliable monitoring and quality assurance in industrial and logistics environments.

---

🧠 Main Flow – Inference and Counting Engine

The Main Flow is the operational core that handles camera input, model inference, and counting logic.
It integrates multiple cameras, model-based detection, and rule-based aggregation to deliver accurate product counts in real time.

Components

• Camera Input:
  Supports one or several camera feeds, including IP cameras or RTSP/RTMP streams positioned along the conveyor.
  Each camera can be independently configured for frame rate, field of view, and detection zones.

• Inference Module:
  Uses a detection model (e.g., YOLO, EfficientDet, or custom-trained object detector) to identify products as they pass through the camera’s region of interest.
  Models can detect multiple product types and assign labels accordingly.

• Counting Logic:
  Once detections are made, products are tracked across consecutive frames to ensure accurate counting without duplication.
  Adjustable rules allow for:

  • Direction-based counting (entry vs. exit).
  • Minimum detection area thresholds.
  • Temporal filtering to avoid false positives.

• Configuration Settings:
  Users can modify:

  • Model parameters (thresholds, classes to detect).
  • ROI definitions for each camera.
  • Counting sensitivity and debounce intervals.
  • Triggered actions (e.g., capture image on detection).

Outputs

• Real-time count per class and camera.
• Aggregated statistics for session summaries.
• Triggered captures sent to the Capture and Sessions modules.

This flow can be adapted for packaging lines, assembly stations, or sorting conveyors, where precision and traceability are critical.

---

🖥️ Monitor – Live Visualization and Status Control

The Monitor subflow provides an operational interface to observe counting activity in real time and control the system.

Features

• Live video display for each active camera, with visual overlays showing detection boxes and current counts.
• Per-camera metrics such as frame rate, detection latency, and product throughput.
• Model performance visualization, including detection confidence distribution.
• Start, stop, and reset session controls to manage counting runs.
• Alerts and system logs for missed frames, detection errors, or over-threshold activity.

Monitor functions as the central control panel for supervising production in real time.

---

📸 Capture – Automatic Image Dataset Builder

The Capture module is responsible for saving frames or cropped images from product detections into structured datasets.
It mirrors the functionality of the Training module from the Generic Multicam Flow but is tailored for product counting use.

Capabilities

• Automatic image capture whenever a new object is detected or classified.
• Selective capture based on product label, confidence threshold, or custom triggers.
• Folder-based organization by camera ID, product type, and timestamp.
• Dataset-ready format for retraining or quality audits.
• Manual capture option from the Monitor interface for operator validation.

This ensures every counted product can be visually verified, creating a valuable dataset for later retraining or traceability purposes.

---

📊 Sessions – Counting History and Review

The Sessions module records every counting session initiated through the Monitor, storing detailed results and allowing post-process supervision.

Functionality

• Session history view: Lists all past runs with metadata such as start time, duration, cameras involved, and total counts per class.
• Result explorer: Displays all individual captured images, organized by label or product type.
• Image preview and verification: Operators can review samples to ensure the counting accuracy of each label.
• CSV export: Download complete session results, including timestamps, detection classes, and camera identifiers.
• ERP integration: Exposes API endpoints for seamless synchronization with production control systems or factory management software.

This module transforms raw detections into actionable business insights and auditable production reports.

---

🧱 Interconnection Between Subflows

Subflow	Purpose	Main Interactions
Main Flow	Core inference and counting engine	Sends detections and statistics to Monitor, Capture, and Sessions
Monitor	Supervision and control	Starts, pauses, and monitors Main Flow sessions
Capture	Frame storage and dataset creation	Receives trigger images from Main Flow
Sessions	Results history and ERP integration	Aggregates data from all modules, enables verification and export

---

💡 Typical Use Cases

• Counting bottles, boxes, or components passing along industrial conveyors.
• Automated verification of packaging processes.
• Sorting and labeling systems with per-class statistics.
• Dataset creation for visual inspection model improvement.
• Integration with ERP systems for production tracking and quality control.

---

🧾 Summary

The Product Counting Flow by Rosepetal AI is a purpose-built flow for industrial automation and smart manufacturing scenarios.
It combines multi-camera vision, AI-based detection, and structured session management to deliver real-time product counting with full traceability — from live monitoring to labeled image review.

---