Shelf Monitoring IoT Retail — Implementation & Analysis

Abstract — Shelf Monitoring IoT Retail

Shelf Monitoring Iot Retail Comprehensive Technical Investigation & Design Architecture · Prepared for Research & Evaluation 1. Executive Overview & Research Problem Statement This technical report presents the comprehensive implementation of a real-time retail shelf monitoring system utilizing computer vision and machine learning technologies. The system leverages existing CCTV infrastructure to automatically detect out-of-stock (OOS) situations and product misplacements through advanced video analysis techniques.

Operational Methodology & Framework

The system aims to monitor retail shelves using existing CCTV cameras, providing real-time analysis and alerts for: - Out-of-stock (OOS) detection - Product misplacement identification The system follows Clean Architecture principles with clear separation of concerns: - Entities : Core business objects (SKU, Planogram, Detection, Alert) - Use Cases : Business logic implementation - Adaptors : External service interfaces (ML models, databases, tracking) - Frameworks : Infrastructure and UI components Keyframe Selection : - Intensity-based change detection to identify significant frames - Reduces computational load by processing only meaningful frames - Non-keyframes utilize tracking algorithms (SORT) for continuity Indexing Phase (One-time setup per shelf): 1. Capture reference shelf images with correct product placements 2. Run YOLOv12 detection on reference images 3. Apply clustering algorithm (DBSCAN) for row identification: - Cluster bounding boxes by Y-coordinate - Sort rows top-to-bottom by average Y-coordinate - Order items within rows left-to-right by X-coordinate 4. Generate grid matrix structure: {shelf_id, rows: [{row_idx, items: [{item_idx, bbox, sku_id}]}]} 5. Store planogram in PostgreSQL database Runtime Processing : 1. Apply same clustering algorithm to current frame detections 2. Match current grid structure to reference planogram 3. Compare detected SKUs with expected SKUs per cell position 4. Generate cell states: OK , OOS , MISPLACED , UNKNOWN Threading Benefits : - Decoupled detection from grid analysis and alert generation - Faster video rendering with real-time overlay updates - Improved system responsiveness and throughput Challenge : Real-time processing requirements with limited computational resources Figure 1: Conceptual Model & Experimental Workflow for Shelf Monitoring Iot Retail 2. System Architecture & Functional Block Diagram Solutions : - Hardware-accelerated inference (OpenVINO, TensorRT) - Keyframe-based processing to reduce computational load - Multi-threaded architecture for parallel processing - Efficient tracking algorithms (SORT) for non-detection frames The Retail Shelf Monitoring System successfully demonstrates the integration of modern computer vision and machine learning technologies for practical retail automation. The system achieves real-time performance through strategic optimization techniques while maintaining high accuracy through robust algorithmic approaches.

Experimental Findings & Validation

A real-time retail shelf monitoring solution powered by computer vision and machine learning. The system leverages existing CCTV infrastructure to automatically detect out-of-stock (OOS) situations and product misplacements , enabling data-driven shelf management and faster restocking decisions.

Scope & Practical Applications

• Real-time automated control and signal telemetry
• Embedded verification testbeds for institutional research
• High-throughput data streaming and anomaly alerting
• Performance optimization and resource consumption minimization
• Full integration ready for IEEE dissertation submission