🏆 Presented at ICNCS 2025, VIT Chennai Hybrid LSTM + YOLOv8 system for predictive demand forecasting and real-time smart shelf monitoring
- Overview
- Key Features
- System Architecture
- Tech Stack
- Project Structure
- Getting Started
- Model Details
- Results
- Pipeline Workflow
- Conference
- Author
IntelliStock is an intelligent, hybrid inventory management system that combines the power of deep learning-based time series forecasting (LSTM/BiLSTM) with real-time computer vision (YOLOv8) to automate stock monitoring and predict demand before shortages occur.
Traditional inventory systems are reactive — they restock after products run out. IntelliStock flips this model by:
- 📷 Seeing shelf stock levels in real-time via camera feeds using YOLOv8
- 📈 Predicting future demand using LSTM-based sequence modeling
- 🔔 Alerting when to reorder, before shelves go empty
This makes it ideal for retail stores, warehouses, supermarkets, and smart supply chains.
- 🎯 Hybrid AI Pipeline — Seamlessly integrates YOLOv8 object detection with LSTM demand forecasting
- 📊 Predictive Refill Alerts — Forecasts stock depletion before it happens
- 🖼️ Image-Driven Inventory Tracking — Real shelf images → automatic item counts
- ⚡ Scalable Batch Pipelines — Efficiently processes large-scale image and time-series data
- 📉 Ultra-Low Forecasting Error — Achieved MAE = 0.0020 on test data
- 🔄 End-to-End Automation — From image capture to reorder recommendation, fully automated
- 📦 Modular Design — Vision and forecasting modules can be used independently
┌─────────────────────────────────────────────────────────┐
│ IntelliStock Pipeline │
│ │
│ 📷 Camera / Image Feed │
│ │ │
│ ▼ │
│ ┌─────────────────┐ ┌───────────────────────────┐ │
│ │ YOLOv8 Module │────▶│ Stock Count Extractor │ │
│ │ (Object Det.) │ │ (ROI-based Counting) │ │
│ └─────────────────┘ └───────────┬───────────────┘ │
│ │ │
│ ┌───────────▼───────────────┐ │
│ │ Time Series Database │ │
│ │ (Stock Level History) │ │
│ └───────────┬───────────────┘ │
│ │ │
│ ┌───────────▼───────────────┐ │
│ │ LSTM / BiLSTM Forecaster │ │
│ │ (Demand Prediction) │ │
│ └───────────┬───────────────┘ │
│ │ │
│ ┌───────────▼───────────────┐ │
│ │ Alert & Reorder Engine │ │
│ │ 📦 Refill Recommendation │ │
│ └───────────────────────────┘ │
└─────────────────────────────────────────────────────────┘
| Category | Tools |
|---|---|
| Computer Vision | YOLOv8 (Ultralytics), OpenCV |
| Deep Learning | PyTorch, LSTM, BiLSTM |
| Data Processing | Pandas, NumPy |
| Visualization | Matplotlib, Seaborn |
| ML Utilities | Scikit-learn |
| Deployment | Flask / Streamlit |
| Version Control | Git, GitHub |
| Language | Python 3.9+ |
IntelliStock/
│
├── 📂 data/
│ ├── raw/ # Raw image datasets & sales records
│ ├── processed/ # Cleaned & preprocessed data
│ └── time_series/ # Stock level time series data
│
├── 📂 models/
│ ├── yolov8/ # YOLOv8 weights & config
│ │ ├── best.pt # Trained YOLOv8 model
│ │ └── data.yaml # Dataset config
│ └── lstm/
│ ├── lstm_model.pth # Trained LSTM weights
│ └── bilstm_model.pth # Trained BiLSTM weights
│
├── 📂 src/
│ ├── detection/
│ │ ├── detect_shelf.py # YOLOv8 shelf detection
│ │ └── roi_extractor.py # ROI-based stock counter
│ ├── forecasting/
│ │ ├── lstm_model.py # LSTM architecture
│ │ ├── bilstm_model.py # BiLSTM architecture
│ │ └── train.py # Training script
│ ├── pipeline/
│ │ ├── batch_pipeline.py # Scalable batch processing
│ │ └── alert_engine.py # Reorder alert system
│ └── utils/
│ ├── preprocess.py # Data preprocessing
│ └── visualize.py # Plotting & dashboards
│
├── 📂 notebooks/
│ ├── EDA.ipynb # Exploratory data analysis
│ ├── LSTM_Training.ipynb # Model training walkthrough
│ └── YOLOv8_Training.ipynb # Vision model training
│
├── 📂 results/
│ ├── metrics/ # MAE, RMSE, accuracy logs
│ └── plots/ # Forecast charts & detection outputs
│
├── app.py # Streamlit / Flask app entry point
├── requirements.txt
└── README.md
Python >= 3.9
pip
Gitgit clone https://github.com/Akhila707/Intellistock.git
cd Intellistockpython -m venv venv
source venv/bin/activate # Linux/Mac
venv\Scripts\activate # Windowspip install -r requirements.txtpython src/detection/detect_shelf.py --source data/raw/shelf_images/python src/forecasting/train.py --epochs 100 --model lstmpython src/pipeline/batch_pipeline.pystreamlit run app.py| Parameter | Value |
|---|---|
| Base Model | YOLOv8n / YOLOv8s |
| Task | Object Detection |
| Input | Shelf camera images |
| Output | Bounding boxes + item counts |
| Training Data | Custom shelf image dataset |
- Detects individual products on retail shelves
- Counts items per shelf zone using ROI (Region of Interest) extraction
- Feeds real-time stock count data into the forecasting pipeline
| Parameter | Value |
|---|---|
| Architecture | LSTM / Bidirectional LSTM |
| Input | Sequential stock level history |
| Output | Predicted future demand |
| Loss Function | Mean Absolute Error (MAE) |
| Test MAE | 0.0020 |
| Optimizer | Adam |
- Learns temporal patterns in stock depletion
- BiLSTM captures both forward and backward dependencies in demand patterns
- Outputs future demand values to trigger reorder alerts
| Metric | Value |
|---|---|
| MAE (Mean Absolute Error) | 0.0020 |
| Detection Accuracy | High precision on shelf items |
| Pipeline Scalability | Batch processing ready |
| Alert Lead Time | Configurable (hours / days ahead) |
📉 MAE of 0.0020 achieved on demand forecasting test set — indicating near-perfect stock prediction accuracy.
1. 📷 Capture shelf images (live camera / batch upload)
│
2. 🔍 YOLOv8 detects & counts products per shelf zone
│
3. 📥 Stock counts logged to time series database
│
4. 🧠 LSTM/BiLSTM forecasts next N-day demand
│
5. ⚠️ Alert engine checks forecast vs. reorder threshold
│
6. 📦 Reorder recommendation generated automatically
This project was presented at:
ICNCS 2025 — International Conference on Networks and Communication Systems Venue: VIT Chennai Paper: IntelliStock — Predictive Refill & Smart Shelf Monitoring Author: PV Akhila
PV Akhila Data Scientist | AI & ML Researcher | Physics Background
⭐ If you find this project useful, please consider giving it a star!