Online-Learning Discriminator Selection for Stable Subject Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The accuracy of tracking using online learning decreases when there is insufficient training data or a small number of training sessions, leading to instability and erroneous tracking.
Innovation Solution
An information processing apparatus with multiple discriminators, including a first discriminator that performs online learning, selectively uses either or both discriminators based on the completeness of online learning, integrating likelihood maps with weighted sums according to completeness evaluations to suppress errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If online learning is used to improve tracking accuracy, then tracking accuracy is improved when sufficient training data is available, but tracking accuracy decreases and instability increases when training data is insufficient
Solution Approach 1:
The patent dynamically adjusts the learning state of the discriminator based on training data availability. When training data is sufficient, the system performs online learning to improve tracking accuracy. When training data is insufficient, the system switches to a fixed learning state to maintain stability, thus adapting the tracking system's behavior to current data conditions.
Solution Approach 2:
The patent changes the learning parameter (learning rate or training mode) of the discriminator based on the completeness evaluation. The determination unit switches between online learning mode and fixed mode by adjusting the learning parameter, allowing the system to optimize tracking accuracy when data is abundant while maintaining reliability when data is limited.
2Adaptability or versatility
If a single discriminator performs online learning, then tracking adaptability is improved, but tracking reliability decreases when online learning is not complete
Solution Approach 1:
The patent segments the tracking system into two distinct discriminators: one performing online learning for adaptability and another maintaining fixed parameters for reliability. The selection unit chooses which discriminator to use based on the completeness evaluation, ensuring that each discriminator performs its specialized function without interfering with the other's performance characteristics.
Solution Approach 2:
The determination unit acts as an intermediary that evaluates the completeness of online learning and mediates between the adaptive discriminator and the reliable discriminator. Based on this evaluation, it selects the appropriate discriminator for tracking, thus bridging the gap between adaptability and reliability requirements.
3Reliability
If multiple discriminators are used to maintain tracking reliability, then tracking stability is improved, but device complexity increases
Solution Approach 1:
The patent designs multiple discriminators with different functions: one discriminator is optimized for online learning and adaptability, while another is optimized for stability and reliability. Each discriminator serves a specific purpose, and the selection unit determines which one to use based on current conditions, thus achieving multi-functionality without requiring a single complex discriminator to handle all scenarios.
Data Source
AI summary
There is provided with an information processing apparatus. A tracking unit tracks a subject in an input image using one or both of a first discriminator that tracks the subject and a second discriminator that tracks the subject and is different from the first discriminator. An obtaining unit obtains training data used for training to track with the first discriminator. A learning unit performs online learning of causing the first discriminator to learn while tracking the subject using the training data. An evaluating unit evaluates a completeness of the online learning. A determination unit determines whether or not the tracking unit is to use the first discriminator to track the subject according to the evaluation of the completeness.