Dynamic Filter Switching for Object Tracking Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object tracking technologies face challenges in accurately tracking targets with high computational load, leading to potential delays in warnings and vehicle control due to ambiguity in velocity detection, which compromises real-time tracking performance.
Innovation Solution
An object tracking apparatus that generates multiple target candidates with assumed velocity ambiguity, uses filters with varying state variables to calculate prediction and estimation values, and selectively switches to filters with more state variables as the number of candidates decreases to balance tracking accuracy and computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple target candidates with velocity ambiguity are generated and tracked using filters, then tracking accuracy is improved, but computational load increases
Solution Approach 1:
The patent dynamically switches between a first filter (simpler, fewer state variables) and a second filter (more complex, more state variables) based on the number of target candidates. When target candidates are numerous, the simpler first filter is used to reduce computational load. When target candidates are few, the more accurate second filter is used to improve tracking precision. This dynamic adaptation resolves the contradiction between accuracy and computational complexity.
Solution Approach 2:
The patent changes the parameter of filter complexity (number of state variables) based on the number of target candidates. The system selects different filter configurations (first filter with fewer state variables vs. second filter with more state variables) according to operational conditions, thereby optimizing the balance between computational efficiency and tracking accuracy in different scenarios.
2Measurement precision
If filters with more state variables are used, then tracking precision is improved, but real-time performance deteriorates due to increased computation time
Solution Approach 1:
The system dynamically adjusts filter complexity based on the number of target candidates. When candidates are numerous, the simpler first filter maintains real-time performance. When candidates are few, the more precise second filter is activated without significantly impacting real-time performance. This dynamic switching resolves the contradiction between precision and real-time performance.
Solution Approach 2:
The patent applies partial action by using the simpler first filter when it provides sufficient tracking precision for the current number of targets. The more complex second filter is only activated when necessary (when target candidates are few), avoiding unnecessary computational overhead and maintaining real-time performance in most operational scenarios.
3Measurement precision
If velocity ambiguity is resolved by tracking multiple target candidates, then velocity identification accuracy is improved, but processing time increases
Solution Approach 1:
The patent dynamically selects filter complexity based on the number of target candidates to optimize the balance between velocity identification accuracy and processing time. The first filter provides faster processing for large numbers of candidates, while the second filter provides more accurate velocity identification when candidates are few, thereby resolving the time-accuracy trade-off.
Data Source
AI summary
An object tracking apparatus generates a plurality of target candidates in which ambiguity in velocity is assumed for an object detected for a first time, calculates a current prediction value of a state quantity of each target candidate from a past estimation value of the state quantity of each target candidate, using one filter among a plurality of filters, and calculates the current estimation value of the state quantity of the target candidate from each of the calculated prediction values and a current observation value that matches each prediction value. The object tracking apparatus deletes the target candidate of which a determined likelihood is less than a preset threshold among the target candidates, and switches a used filter to a filter that has a larger number of state variables among the filters, in response to the target candidate being deleted and the total number of target candidates decreasing.