Extended Object Tracking Using MAP-Based Point Cloud Association
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing extended object tracking systems face challenges in accurately associating multiple noisy sensor measurements from spatially distributed reflection points, especially for objects occupying multiple sensor cells, due to the unknown shape and varying nature of these objects over time, making it difficult to consider all association possibilities efficiently.
Innovation Solution
A sensing apparatus and method that utilize marginal association probabilities (MAPs) to determine associations between measurement points and objects, followed by a density-based spatial clustering algorithm (DBSCAN) to form groups and classify objects, while excluding high MAP threshold values to improve tracking accuracy, especially for closely spaced targets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all association possibilities are considered for extended objects generating multiple measurements, then tracking accuracy is improved, but computational complexity becomes infeasible for increasing number of targets and measurements
Solution Approach 1:
The patent segments the complex association problem by introducing a two-stage process: first performing point cloud gating to pre-filter measurement points based on spatial consistency with predicted object positions, then performing association only within the gated subsets. This segmentation divides the intractable global association problem into manageable local association problems, maintaining tracking accuracy while reducing computational complexity from considering all possible associations to only considering gated associations.
2Measurement precision
If point cloud gating is performed before association for extended objects, then association accuracy is improved, but the gating process adds computational overhead
Solution Approach 1:
The patent applies preliminary action by performing point cloud gating before the association step. The gating process pre-filters measurement points by checking their spatial consistency with predicted object positions using gates based on object state predictions. This preliminary filtering removes obviously inconsistent measurements before they enter the association process, improving association accuracy while the computational overhead is justified by the reduction in subsequent association computations.
Data Source
Figure 1~2
Figure 3a~3c
Figure 4~5
AI summary
An apparatus (100) and method (600) for tracking one or more extended objects (110a-c) are disclosed. The apparatus (100) comprises at least one sensor (107) configured to perform a plurality of measurements of the one or more extended objects (110a-c) for obtaining a point cloud of measurement points (601). Moreover, the apparatus (100) comprises a processing circuitry (101) configured to determine for each measurement point of the point cloud and each object of the one or more extended objects (110a-c) a respective marginal association probability, MAP, value of the respective measurement point for being associated with the respective object (603). Furthermore, the processing circuitry (101) is configured to define for each object of the one or more objects (110a-c) a respective subset of the point cloud of measurement points based on the plurality of MAP values (605) and to assign one or more measurement points of the subset of the point cloud to the respective object of the one or more extended objects (110a-c) (607).