Image-Based Object Interaction Inference Using Social Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems struggle to effectively infer relationships between objects in images, particularly in situations like searching for missing persons with disabilities or dementia, where traditional methods fail to recognize significant interactions.
Innovation Solution
A device and method utilizing image recognition to extract attributes, calculate trajectories, and select main objects interacting with a target object through interaction indices, generating a social graph to identify key individuals in images from multiple sources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image recognition methods are used to identify objects in images, then basic object detection can be achieved, but the ability to infer interaction relationships between objects is insufficient
Solution Approach 1:
The system segments the interaction inference process into distinct components: trajectory calculation for movement patterns, attribute extraction for object characteristics, and interaction index calculation for relationship assessment. This segmentation allows each component to be optimized independently while collectively improving interaction relationship inference accuracy without overwhelming system complexity
Solution Approach 2:
The patent introduces a new dimension of analysis by calculating interaction indices that combine spatial distance, temporal co-occurrence, and trajectory overlap. This multi-dimensional approach transforms traditional 2D image recognition into a 3D spatiotemporal analysis, enabling accurate inference of interaction relationships that cannot be detected by conventional methods
2Loss of information
If multiple imaging devices are used to capture images from different locations, then more comprehensive data is obtained, but the complexity of processing and integrating data from multiple sources increases
Solution Approach 1:
The system merges data from multiple imaging devices by establishing a unified coordinate system and integrating trajectories across different camera views. The interaction index calculation combines spatial, temporal, and movement pattern data into a single comprehensive metric, simplifying the integration of multi-source data while maintaining information completeness
Solution Approach 2:
The patent introduces an intermediary processing layer that standardizes data from different imaging devices before analysis. This intermediary layer normalizes coordinates, timestamps, and detection formats, enabling seamless integration of multi-device data without requiring complex custom integration logic for each device pair
3Measurement precision
If detailed interaction attributes are calculated between objects, then more accurate relationship inference is achieved, but the computational time and resources increase
Solution Approach 1:
The system calculates interaction indices using a selective subset of attributes based on the specific search context. Rather than computing all possible interaction metrics for every object pair, the system prioritizes the most relevant attributes (such as spatial proximity and temporal co-occurrence) to achieve accurate relationship inference within acceptable processing time
Solution Approach 2:
The patent performs preliminary filtering of object pairs based on basic criteria (such as minimum distance threshold or time window) before calculating detailed interaction attributes. This preliminary action reduces the number of candidate pairs requiring intensive computation, significantly decreasing processing time while maintaining inference accuracy for relevant cases
Data Source
AI summary
Disclosed is a device and method for inferring a correlation between objects through image recognition. The device for inferring a correlation between objects through image recognition according to an embodiment comprises a communicator and an interaction inferencer configured to select a main object interacting with a target object at a predetermined distance in the input image or generate a social graph including the main object and the target object.