Scene Detection Confidence Update via Transition Probabilities
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current scene detection methods in wireless communication devices, such as smartphones and tablets, are prone to detection errors and are complex to implement, especially in terms of memory size and learning phases, despite the use of meta classifiers and meta filters that exploit temporal correlations.
Innovation Solution
A method that updates the confidence probability of detected scenes based on user habits and transition probabilities between scenes, refining scene detection by weighting initial confidence probabilities with transition probabilities and normalizing them, thereby reducing noise and improving filtering performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If meta classifiers and meta filters are used to improve scene detection reliability, then detection accuracy is improved, but device complexity and memory size increase
Solution Approach 1:
The patent changes the parameter of confidence probability by updating it based on transition probabilities between scenes. Instead of using complex meta classifiers, the system updates the confidence probability of detected scenes by multiplying the initial confidence by a transition probability factor, thereby improving reliability through parameter optimization rather than structural complexity
Solution Approach 2:
The patent performs preliminary action by pre-defining transition probabilities between different scenes before actual scene detection occurs. These transition probabilities are stored in a lookup table and used during runtime to quickly update confidence probabilities without requiring complex real-time computations, thus reducing device complexity while maintaining reliability
2Measurement precision
If transition probabilities are calculated and stored for all scene pairs, then detection accuracy is improved, but memory size increases
Solution Approach 1:
The patent optimizes the storage of transition probabilities by storing them in a compact format. The transition probabilities are stored as pre-computed values in a lookup table with dimensions corresponding to the number of scenes, allowing efficient retrieval during scene detection without storing redundant information. This reduces memory requirements while maintaining the precision needed for accurate scene detection
3Reliability
If confidence probability is updated using transition probabilities, then noise is reduced and filtering performance is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary computation of transition probabilities and stores them in a lookup table. During runtime, the system simply retrieves the pre-computed transition probability from the table and multiplies it with the initial confidence probability, avoiding complex real-time calculations. This preliminary action significantly reduces computational complexity while maintaining filtering performance
Solution Approach 2:
The patent implements a feedback mechanism where the updated confidence probability is fed back into the scene detection system. The confidence probability is updated by multiplying the initial confidence by the transition probability, and this updated value is used by the meta-filter to make filtering decisions. This feedback loop improves filtering performance by continuously refining the confidence assessment based on scene transition patterns
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
Method of controlling scene detection by a device (APP) from among a set of possible reference scenes, comprising assigning an identifier to each reference scene, detecting scenes from said set of possible reference scenes at successive detection times using at least one classification algorithm, and filtering (10) these detected current scenes from the identifier (ID) of each new detected current scene and a confidence probability (PC1) associated with this new detected current scene, said confidence probability (PC1) being updated according to a first transition probability (TrPTl) from a previously detected scene to the new detected current scene, the output of the filtering process successively delivering filtered detected scenes (15).