Robust Shelf Monitoring Using Supervised Learning for Improving On-Shelf Availability in Retail Stores

Area of Science:

• Computer Science
• Artificial Intelligence
• Retail Operations

Background:

• High on-shelf availability is crucial for retail profitability.
• Accurate monitoring of product stock levels is essential for maintaining availability.
• Existing methods may lack robustness in detecting subtle product changes on shelves.

Purpose of the Study:

• To develop a robust method for monitoring retail shelves using supervised learning.
• To improve on-shelf availability by accurately detecting product increases and decreases.
• To provide a system that assists store clerks in maintaining optimal stock levels.

Main Methods:

• Image processing techniques including background subtraction and moving object removal to detect changed regions.
• Supervised learning with convolutional neural networks to classify product changes (e.g., taken, replenished).
• Updating shelf conditions and calculating visible product amounts for on-shelf availability.

Main Results:

• Demonstrated effectiveness of product change classification in initial experiments.
• Achieved an 89.6% success rate for on-shelf availability with a one-product error margin.
• Validated the method using real-world video data from a retail store.

Conclusions:

• The proposed method accurately monitors retail shelves and detects product stock changes.
• High accuracy in on-shelf availability monitoring can lead to increased retail profits.
• This approach offers a practical solution for store clerks to maintain optimal product availability.