Learning Model Selection for Object Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing environment recognition methods using sensors often acquire only partial information, making it difficult to accurately identify and distinguish multiple overlapping objects in a real-space environment, especially when infinite candidate objects are present and they overlap or contact each other.
Innovation Solution
An information processing apparatus and method that employs a learning model-based system to select the appropriate sensing settings by calculating latent variable probability distributions and information amounts from sensor data, allowing for more accurate recognition of objects by determining the category and label of observed data through generative models and latent variable analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If traditional sensor-based environment recognition methods are used, then the system is simple and easy to operate, but the information amount obtained is reduced and only partial environment information is acquired
Solution Approach 1:
The patent applies preliminary action by pre-storing multiple learning models corresponding to different categories (e.g., character types, objects) before actual recognition occurs. When sensor data is acquired, the system can immediately select and apply the appropriate pre-trained model without needing to process all possible categories in real-time, thus reducing information loss while maintaining manageable system complexity through organized model storage and selection mechanisms
Solution Approach 2:
The patent utilizes parameter changes by varying the selection of learning models based on the specific category of objects being recognized. Different learning models with different parameters are selected depending on the detection target (e.g., different character types, different object categories), allowing the system to adapt to different recognition tasks and maximize information extraction from sensor data without requiring a single overly complex universal model
2Measurement precision
If active sensing with repeated position changes is performed, then more complete environment information is acquired, but the time required for recognition increases
Solution Approach 1:
The patent implements feedback by using the output results from learning models to guide subsequent sensing actions. When a learning model identifies potential objects or categories, the system uses this feedback information to determine the next sensing position or parameters, creating an iterative process that efficiently converges on accurate recognition without requiring exhaustive sampling of all possible positions, thus reducing recognition time while maintaining precision
Solution Approach 2:
The patent applies preliminary action by pre-processing sensor data through multiple learning models before final recognition is made. The system performs preliminary classification and identification using stored models, which narrows down the search space and allows for more targeted subsequent sensing actions, reducing the total time required while improving recognition accuracy through multi-model validation
3Measurement precision
If multiple learning models are selected and processed, then recognition accuracy for overlapping objects improves, but calculation complexity and processing time increase
Solution Approach 1:
The patent applies segmentation by dividing the recognition task into separate processing stages, each handled by different learning models organized by category. Instead of using one monolithic complex model, the system segments the problem into multiple specialized models that process different aspects or categories of objects, making the overall system more manageable while improving accuracy through specialized processing for each segment
Solution Approach 2:
The patent uses partial action by selectively applying only the necessary learning models based on the specific recognition task and sensor data characteristics. Rather than processing through all available models regardless of situation, the system intelligently selects and applies only the relevant subset of models needed for the current detection scenario, reducing processing complexity while maintaining recognition accuracy through targeted model selection
4Measurement precision
If feature variables of sensor data are used directly for work plan drafting, then the process is simple, but accurate recognition of overlapping or contacting objects becomes difficult
Solution Approach 1:
The patent introduces learning models as intermediary components between raw sensor data and final recognition decisions. These intermediary models process and interpret sensor data, extracting meaningful features and patterns that directly feature variable processing would miss, especially for overlapping objects. This intermediary layer adds controlled complexity that significantly improves object distinction accuracy by providing intelligent interpretation of complex sensor inputs
Data Source
AI summary
An information processing apparatus and an information processing method capable of accurately recognizing an object to be sensed are provided.One or a plurality of learning models are selected from among learning models corresponding to a plurality of categories, a priority of each of the selected learning models is set, observation data obtained by sequentially compounding pieces of sensor data applied from a sensor is analyzed using the learning model and the priority of the learning model, a setting of sensing at a next cycle is selected based on an analysis result, predetermined control processing is executed in such a manner that the sensor performs sensing at the selected setting of sensing, and the learning model corresponding to the category estimated as the category to which a current object to be sensed belongs with a highest probability is selected based on the categories to each of which the previously recognized object to be sensed belongs.