3D Object Recognition with Spatial Property Database Updating
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition systems struggle with recognizing objects in images and require time-consuming database construction for each recognition task, often failing to account for relationships among objects.
Innovation Solution
An information processing apparatus that includes an object property group information input unit, an object disposition property database, and a prediction unit to predict and update object properties based on input information, utilizing a neural network to infer unknown or incorrect object properties from surroundings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image recognition is performed using traditional methods, then object type and position can be identified, but objects are not recognized in some cases and database construction is time-consuming
Solution Approach 1:
The system performs preliminary actions by pre-establishing a general object disposition property database that stores spatial relationships between object types before specific recognition tasks. This pre-built database allows the system to quickly infer object properties during actual recognition without time-consuming database construction for each task, thereby reducing time loss while maintaining high recognition accuracy through the use of pre-learned spatial dispositions
2Reliability
If traditional image recognition is used, then processing can be completed, but it fails to account for relationships among objects
Solution Approach 1:
The system applies universality by creating a single object disposition property database that stores general spatial relationships between multiple object types. This universal database serves multiple recognition tasks simultaneously, allowing the system to account for relationships among objects without increasing complexity for each specific task. The database structure and retrieval mechanisms remain consistent across different recognition scenarios, maintaining reliability while avoiding the complexity of task-specific relationship models
3Measurement precision
If object property information is input to the prediction unit, then object labels can be predicted and corrected, but the system requires structured input of object property groups
Solution Approach 1:
The system applies self-service by automatically generating the required object property group information from input images or three-dimensional models. The object property group information input unit extracts object type, position, and spatial relationship data directly from the input data, eliminating the need for manual preparation of structured input. This automatic extraction process maintains high object label accuracy through precise property identification while significantly easing the operational burden of data preparation
Data Source
AI summary
An information processing apparatus includes an object property group information input unit configured to input object property group information that includes at least two or more pieces of object property information, the object property information including type information that represents names of types of objects and information regarding three-dimensional positions of the objects in a space; an object disposition property database that holds the type information that represents the names of the types of the objects and the information regarding the three-dimensional positions of the objects in the space in an associated manner; a prediction unit configured to predict information related to the information included in the object property group information on the basis of the object property group information input by the object property group information input unit and the object disposition property database; and an object disposition property database updating unit configured to update the object disposition property database on the basis of the object property group information input by the object property group information input unit and a result of the prediction of the prediction unit.