Recognizer Relearning Using Reverse-Time Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recognizers used for object recognition, such as those identifying people or cars, are prone to repetitive erroneous recognition when similar conditions arise, leading to a desire for improved performance to prevent such errors.
Innovation Solution
A learning model generation method that tracks a target object in reverse time series, performs relearning based on tracking results, and updates the recognizer using added labels to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a recognizer learned in advance is used for object recognition, then recognition processing can be performed efficiently, but erroneous recognition occurs repeatedly when similar cases arise
Solution Approach 1:
The system implements feedback by tracking recognized objects across multiple frames in reverse chronological order, identifying cases where recognition results change over time. This feedback mechanism detects erroneous recognitions and triggers relearning processes to improve future recognition accuracy while maintaining efficient processing through the use of pre-trained models for initial recognition.
Solution Approach 2:
The system performs preliminary actions by pre-training recognizers with large datasets before deployment. Additionally, it proactively tracks and identifies potential erroneous recognition cases before they propagate, using reverse-time tracking to detect patterns of repeated errors and initiate corrective relearning processes in advance.
2Reliability
If relearning is performed using forward time series tracking, then recognition results can be updated, but it is difficult to identify repetitive erroneous recognition patterns
Solution Approach 1:
The system inverts the conventional forward-time tracking approach by performing tracking in reverse chronological order. This inversion allows the system to trace back from current recognition results to historical frames, efficiently identifying patterns where recognition results change over time and detecting repetitive erroneous recognition that would be difficult to capture with forward-only tracking.
Solution Approach 2:
The system maintains continuous tracking of recognized objects across multiple frames in reverse time series, ensuring uninterrupted monitoring of recognition patterns. This continuous reverse tracking enables the system to consistently identify error patterns and maintain reliable recognition accuracy over extended periods without losing information about temporal relationships.
Data Source
AI summary
The present technology relates to a learning model generation method, an information processing device, and an information processing system capable of constructing a recognizer with less erroneous detection. A target object is tracked in a reverse direction in time series, the target object being recognized by recognition processing using a recognizer to which a learning model for performing recognition processing on input data is applied, and relearning of the learning model is performed by using data generated on the basis of a result of the tracking. The data is generated by tracking the target object in the reverse direction in time series and adding a label to the target object tracked. The present technology can be applied to, for example, an information processing device that performs relearning of a recognizer that recognizes a predetermined object.