Video Interaction Analysis for Customer-Product Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sales strategies often fail to effectively increase customer motivation to purchase products due to reliance on individual sales staff skills, lacking targeted information based on customer-product interactions.
Innovation Solution
An information processing system utilizing cameras and display devices to analyze customer-product interactions, employing machine learning models to provide relevant product information based on identified relationships, thereby enhancing sales staff assistance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If video analysis and machine learning models are implemented to identify customer-product interactions and provide targeted information, then customer motivation to purchase is improved, but device complexity and processing requirements increase
Solution Approach 1:
The system segments the retail environment into multiple monitoring zones with specific analysis functions. Different machine learning models are assigned to different interaction types (looking, touching, grasping), allowing the complex analysis task to be divided into manageable segments that can be processed independently and efficiently.
Solution Approach 2:
Machine learning models are pre-trained offline to recognize various customer-product interaction patterns. This preliminary action enables the system to quickly classify interactions during runtime without performing complex training in real-time, reducing processing complexity while maintaining high accuracy in identifying customer intentions.
2Loss of time
If real-time video analysis is performed to detect customer-product interactions, then relevant product information can be provided timely, but processing time and computational resources increase
Solution Approach 1:
The system applies different analysis depths to different zones based on detected interaction types. For example, simple 'looking' interactions receive basic model matching, while more complex 'grasping' interactions trigger more detailed analysis. This localized quality adjustment optimizes computational resource allocation according to the specific analysis needs of each situation.
Solution Approach 2:
The system performs partial analysis by first identifying basic interaction types using lightweight models, then selectively applying more computationally intensive analysis only when necessary (e.g., when customer intent is ambiguous or high-value products are involved). This approach reduces overall processing time and resource consumption while maintaining accuracy for critical cases.
3Measurement precision
If multiple machine learning models are used to analyze different interaction types, then measurement precision of customer intent is improved, but model selection and management complexity increase
Solution Approach 1:
A universal model selection mechanism is implemented that automatically chooses the appropriate machine learning model based on the detected interaction type. This single multi-functional selection system handles all interaction types (looking, touching, grasping, and other relationships) without requiring separate management systems for each model, reducing overall management complexity while maintaining high detection accuracy.
Solution Approach 2:
An intermediary model selection layer is introduced between video input and analysis models. This intermediary component translates diverse interaction types into standardized model selection criteria, acting as a mediator that simplifies the complexity of managing multiple specialized models by providing a unified interface for model selection and management.
Data Source
AI summary
A non-transitory computer-readable recording medium has stored therein an information processing program that causes a computer to execute a process including acquiring a video analyzing the acquired video identifying, based on a result of the analyzing, a first area containing a first object included in the acquired video, a second area containing a second object included in the acquired video, and a relationship that identifies interaction between the first object and the second object based on the specified relationship, selecting a model that is relevant to any one of the first object and the second object from a plurality of models and outputting the selected model.