Selected Target Tracking Using Single- and Multi-Target State Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vision-based single tracking algorithms, whether depth learning-based or traditional machine learning-based, suffer from poor robustness when dealing with similar interference targets of the same category, especially when they are obstructed.
Innovation Solution
A target tracking method that incorporates both single-target and multiple-target tracking results to determine state information, using techniques such as intersect over union (IOU) and category information to improve robustness by distinguishing between obstructed and unobstructed targets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If single-target tracking algorithms are used, then tracking speed is maintained, but robustness against similar interference targets deteriorates
Solution Approach 1:
The patent merges single-target tracking results with multiple-target tracking results by combining their respective confidence values and position information. This integration allows the system to leverage the speed advantage of single-target tracking while incorporating the robustness of multiple-target tracking to distinguish the selected target from similar interference targets.
Solution Approach 2:
The system uses feedback from multiple-target tracking to adjust and refine the single-target tracking results. By continuously comparing the selected target's position and confidence with other detected targets, the system can identify and correct tracking errors caused by similar interference targets.
2Reliability
If multiple-target tracking results are integrated, then robustness against interference improves, but system complexity increases
Solution Approach 1:
The patent applies partial action by selectively using only the necessary components from multiple-target tracking (position information and confidence values) rather than implementing the full multiple-target tracking pipeline. This approach improves robustness against interference while avoiding the excessive computational complexity of complete multiple-target tracking.
Solution Approach 2:
The system segments the tracking process into independent components: single-target tracking module, multiple-target tracking module, and result integration module. Each module operates independently and contributes specific information, allowing the system to achieve improved robustness while maintaining manageable complexity through modular design.
Data Source
AI summary
A target tracking method, apparatus, device, and storage medium are disclosed. A single-target tracked result for a selected target and a multiple-target tracked result for multiple targets are acquired. The multiple targets include the selected target which is a target determined according to position information from an external device. The state information about the selected target is determined according to the single-target tracked result and the multiple-target tracked result. The selected target is tracked according to the state information about the selected target, the single-target tracked result and the multiple-target tracked result, so that the robustness of single-target tracking can be improved is achieved.