Planogram Compliance Using Scale-Aware Shelf Image Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in accurately detecting and managing product placement on retail shelves to ensure planogram compliance due to variations in image quality, illumination, resolution, and product size, particularly when dealing with similar products of different sizes and orientations.
Innovation Solution
A system and method that utilizes photometric and geometric transformations, feature extraction, and a trained model to identify product centers and scales, followed by non-maximal suppression to resolve overlaps, generating a realogram for compliance reporting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single marketing image template is used for product detection in uncontrolled retail environments, then the detection process is simplified, but the accuracy deteriorates due to variations in image quality, illumination, and resolution
Solution Approach 1:
The system performs photometric and geometric transformations on the single marketing image template beforehand to generate multiple augmented images that simulate various retail environment conditions (different illuminations, resolutions, angles). This preliminary preparation allows the detection system to handle diverse real-world scenarios without increasing operational complexity.
Solution Approach 2:
The system transforms the template image by changing multiple parameters simultaneously including illumination conditions, image resolution, geometric orientation, and scale. These parameter transformations create a robust set of augmented templates that maintain detection accuracy across varying retail environment conditions.
2Reliability
If multiple region proposals are generated for product detection, then the detection coverage is improved, but the computational overhead and false positives increase
Solution Approach 1:
The system replaces the traditional exhaustive region proposal generation mechanism with a direct feature matching approach. By extracting and matching feature points between the augmented template and target image, the system directly identifies product locations without generating numerous region proposals, thereby reducing computational overhead while maintaining detection coverage.
Solution Approach 2:
The system extracts only the essential feature points from the augmented template images that are most indicative of product identity and position. This selective extraction of critical features enables accurate detection with minimal computational processing, avoiding the need to analyze entire image regions.
3Measurement precision
If feature matching is performed between template and target images, then product identification accuracy is improved, but the computational time increases due to multiple transformations and matching operations
Solution Approach 1:
The system performs feature extraction and key point identification on the augmented template images during the offline template preparation phase. This preliminary extraction of invariant features (such as SIFT, SURF, or ORB keypoints) allows the online detection phase to perform only efficient feature matching operations, significantly reducing processing time while maintaining high identification accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Planograms are used to create consistency between store locations, to provide proper shelf space allocation, to improve visual merchandising appeal, and to create product-pairing suggestions. Existing solutions do not have a way to accurately estimate the scale of magnification of the object in the shelf image, so unable to distinguish between size variants of the same product. A system and method for facilitating planogram compliance for inventory management in a retail store have been provided. The scales are calculated with use of a vector convergence technique followed by a center clustering which automatically removes outliers. Initially disclosure comprises calculation of scales and centers, then generation of region proposals using those scales and centers, Next, classification of the regions proposed and generation of similarity scores, and on the basis of similarity scores conflict resolution is performed among overlapped region proposals using non-maximal suppression.