Action Detection in Time-Series Images Using Two-Stage Temporal Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional action time-period detection technologies face difficulties in accurately detecting categories and time periods for actions in time-series images, particularly due to challenges in estimating proper time periods independent of categories.
Innovation Solution
A detection apparatus that uses one or more processors to set time-period candidates, input features from time-series images, and output likelihoods and correction information to accurately detect actions, start times, and finish times by employing an estimation model that corrects time-period candidates based on likelihoods, improving the accuracy of time-period and category detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If time-period candidates are estimated independent of categories through model M1, then the detection process can be simplified and executed efficiently, but the accuracy of time-period detection for each category deteriorates
Solution Approach 1:
The detection process is segmented into two distinct stages: first, model M1 estimates time-period candidates independently of categories to maintain efficiency; second, model M2 performs category-specific refinement of these candidates to improve accuracy. This segmentation allows each model to specialize in its strength without compromising the other.
Solution Approach 2:
The time-period candidates estimated by model M1 serve as an intermediary between the initial independent estimation and the final category-specific detection. These candidates act as a bridge that model M2 refines for each category, enabling both efficiency and accuracy.
2Adaptability or versatility
If a two-model system (M1 and M2) is used to estimate time-period candidates and then evaluate likelihoods, then detection coverage can be improved, but the complexity of the detection system increases
Solution Approach 1:
The system merges the strengths of two different approaches: model M1 provides broad, category-independent time-period estimation, while model M2 adds category-specific refinement. By combining these models in sequence rather than choosing one approach, the system achieves comprehensive detection coverage while managing complexity through clear functional division.
3Speed
If model M1 estimates time-period candidates without category information, then processing speed is maintained, but the ability to detect proper time periods for specific categories is reduced
Solution Approach 1:
Model M1 performs a preliminary estimation of time-period candidates without category information, establishing a foundation that maintains processing speed. This preliminary action is then refined by model M2 with category-specific information, ensuring both speed and accuracy are achieved through staged processing.
Data Source
AI summary
A detection apparatus includes one or more processors. The processors set at least one time-period candidate. The processors input, to a first model that inputs a feature acquired from a plurality of time-series images and the time-period candidate and outputs at least one first likelihood indicating a likelihood of occurrence of at least one action previously determined as a detection target and correction information for acquisition of at least one correction time period resulting from correction of the at least one time-period candidate, the feature and the time-period candidate, and acquire the first likelihood and the correction information output from the first model. The processors detect, based on the at least one correction time period acquired based on the correction information and the first likelihood, the action included in the time-series images and a start time and a finish time of a time period of occurrence of the action.