Target Object Determination via Interactive Feature Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current human-machine interactive technologies face difficulties in accurately determining a target object in images when multiple objects of the same type are present, requiring separate user selection and degrading interactive efficiency, and struggle with rare words in natural language inputs due to insufficient training data and feature representation.
Innovation Solution
A method and apparatus that acquire first feature information from images and second feature information from interactive inputs, including semantic features between objects, and perform joint corrections to determine the target object, using word vector conversions and fusion processing to enhance accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object detection technology is used to determine target objects, then the system can identify objects in images, but it cannot accurately distinguish between multiple objects of the same type, requiring separate user selection operations
Solution Approach 1:
The patent transforms the target object determination problem from simple object detection to a parameter-matching problem. By converting interactive input parameters (voice, text, gestures) into feature vectors and comparing them with object feature vectors, the system achieves precise identification without requiring multiple user selections. This parameter transformation enables the system to distinguish between multiple objects of the same type based on subtle feature differences.
Solution Approach 2:
The patent replaces the mechanical interaction method (multiple user selections and confirmations) with an automated feature-matching system. Instead of requiring users to manually select and confirm objects, the system uses interactive input parameters to automatically determine the target object through feature comparison, significantly improving interactive efficiency while maintaining high accuracy.
2Measurement precision
If fine-grained object detection is used to handle object attributes, then the system can detect specific object characteristics, but it requires a large number of annotations for training and performs poorly on objects absent from training data
Solution Approach 1:
The patent creates a universal feature extraction framework that can handle various object attributes and types without requiring separate training for each specific attribute. The system extracts comprehensive feature vectors that encompass multiple object characteristics, enabling it to detect and determine target objects across different categories and attributes using a single unified model, thereby reducing training complexity and improving generalization to unseen objects.
Solution Approach 2:
The patent introduces feature vectors as an intermediary representation between raw image data and target object determination. Instead of directly detecting and classifying objects based on raw pixels, the system converts image data into feature vectors that capture essential characteristics, then performs determination based on parameter matching. This intermediary representation enables the system to handle diverse object attributes without requiring extensive specific training for each attribute type.
3Productivity
If simple target detection technology is used, then the system can quickly identify objects, but it cannot exactly identify the specific object designated by the user when multiple similar objects are present
Solution Approach 1:
The patent performs preliminary feature extraction and parameter conversion before the actual target determination. By pre-processing the image data into feature vectors and pre-converting interactive input parameters into comparable formats, the system prepares all necessary information in advance. This preliminary action enables the final determination step to be both fast and accurate, as the system only needs to perform a straightforward parameter matching operation rather than complex analysis at the decision point.
Data Source
AI summary
Provided are methods and apparatuses for determining a target object in an image based on an interactive input. A target object determining method acquires first feature information corresponding to an image and second feature information corresponding to an interactive input; and determines a target object corresponding to the interactive input from among objects in the image based on the first feature information and the second feature information.