State Estimation Apparatus for Tracking Objects with Varying Sizes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Particle filtering techniques face challenges in accurately tracking objects with varying sizes or multiple objects of different sizes in moving images, as they often rely on single observation likelihoods, which can lead to inadequate state estimation.
Innovation Solution
A state estimation apparatus and method that obtain and process multiple observations over time intervals, selecting and weighting observation data based on posterior probability distributions to calculate more accurate likelihoods, enabling robust tracking of objects with size variations or multiple objects in moving images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If single observation likelihood is used in particle filtering, then computation is simpler, but tracking accuracy deteriorates for objects with varying sizes or multiple objects
Solution Approach 1:
The patent segments the likelihood calculation process by dividing multiple observation data into separate likelihood calculations. Instead of using a single observation, the system calculates individual likelihoods for each observation and combines them, thereby improving tracking accuracy for objects with varying sizes or multiple objects while maintaining manageable computation through structured segmentation
Solution Approach 2:
The patent applies partial action by selectively using multiple observations rather than all possible observations. The system calculates likelihoods for a subset of observations and combines them, achieving improved accuracy without the full computational burden of processing every possible observation, thus balancing accuracy improvement with computation complexity
2Measurement precision
If multiple observations are processed to improve state estimation accuracy, then tracking precision improves, but computation complexity increases
Solution Approach 1:
The patent merges multiple likelihood calculations into a unified state estimation process. By combining likelihoods from multiple observations through multiplication and integrating with prediction distributions, the system achieves improved state estimation accuracy while maintaining a coherent computational framework that manages complexity through systematic merging rather than separate independent calculations
3Measurement precision
If multiple likelihoods are calculated and combined, then posterior probability distribution accuracy improves, but processing time increases
Solution Approach 1:
The patent implements continuous useful action by integrating multiple likelihood calculations into a seamless Bayesian updating process. The system continuously updates the posterior probability distribution by combining prediction distributions with multiple likelihoods in a unified mathematical framework, achieving accurate posterior probabilities while minimizing processing time through continuous rather than discrete batch processing
Data Source
AI summary
The purpose of the present invention is to provide a state estimation apparatus that appropriately estimates the internal state of an observation target by determining likelihoods from a plurality of observations. An observation obtaining unit of the state estimation system obtains, at given time intervals, a plurality of observation data obtained from an observable event. The observation selecting unit selects a piece of observation data from the plurality of pieces of observation data obtained by the observation obtaining unit based on a posterior probability distribution data obtained at a preceding time t−1. The likelihood obtaining unit obtains likelihood data based on the observation data selected by the observation selecting unit and predicted probability distribution data obtained through prediction processing using the posterior probability distribution data. The posterior probability distribution estimation unit estimates posterior probability distribution data representing a state of the observable event based on the predicted probability distribution data obtained by the likelihood obtaining unit and the likelihood data. The prior probability distribution output unit outputs prior probability distribution data based on the posterior probability distribution data estimated by the posterior probability distribution estimation unit as prior probability distribution data at a next time t+1.