Portable Radar Adaptive Clustering for Over- and Under-Clustered Targets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional radar clustering techniques struggle with accurate object detection due to over-clustering and under-clustering, particularly in scenarios with complex environments and Doppler zero cases, leading to false targets and suboptimal center identification.
Innovation Solution
Adaptive clustering techniques that utilize spatial and non-spatial dimensions, such as Doppler and SNR, to correct cluster sizes, remove outliers, and adjust gate sizes dynamically, ensuring accurate representation of individual objects and stable target tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional clustering techniques (DBSCAN, K-means) are used considering only spatial dimensions, then the clustering process is simple and fast, but object detection accuracy deteriorates due to over-clustering and under-clustering
Solution Approach 1:
The patent extends conventional spatial clustering by incorporating additional dimensions including Doppler dimension and virtual azimuth dimension. This multi-dimensional approach allows the system to distinguish between targets that are close in space but differ in Doppler characteristics or virtual azimuth, thereby preventing over-clustering while maintaining computational feasibility through dimension-based differentiation.
Solution Approach 2:
The patent dynamically adjusts clustering parameters including gate sizes in range, azimuth, and Doppler dimensions based on target characteristics and environmental conditions. By adapting parameters such as Doppler gate size and virtual azimuth resolution according to specific scenarios (e.g., highway vs. urban environments), the system optimizes detection accuracy without requiring overly complex fixed-structure algorithms.
2Reliability
If conventional clustering techniques are used, then computational resources are used efficiently, but false targets such as ghost targets and shadow targets cannot be identified and eliminated
Solution Approach 1:
The patent introduces virtual azimuth as an intermediary dimension that mediates between spatial position and Doppler information. This virtual azimuth dimension acts as a bridge to distinguish genuine targets from false targets (ghost targets, shadow targets) by providing an additional discrimination criterion that does not require excessive computational resources, thus maintaining processing efficiency while improving reliability.
Solution Approach 2:
The patent segments the clustering process into multiple stages: initial spatial clustering, followed by Doppler-based refinement, and finally virtual azimuth-based validation. This segmented approach allows the system to process data in manageable steps, eliminating false targets at each stage without requiring all computational resources to be allocated simultaneously, thereby maintaining productivity while enhancing reliability.
3Measurement precision
If Doppler information is incorporated into clustering, then targets with zero Doppler (moving perpendicular to radar) can be detected, but the clustering algorithm becomes more complex
Solution Approach 1:
The patent addresses the Doppler zero case by introducing the virtual azimuth dimension as an additional discrimination axis. When Doppler information alone is insufficient (zero Doppler case), the virtual azimuth dimension provides the necessary differentiation capability to detect and cluster targets moving perpendicular to the radar, maintaining detection accuracy without requiring fundamental changes to the core clustering algorithm.
4Measurement precision
If adaptive gate sizing is implemented to optimize clustering parameters, then clustering accuracy improves, but computational overhead increases
Solution Approach 1:
The patent implements adaptive gate sizing that dynamically adjusts clustering parameters based on detected target characteristics and environmental conditions. Gate sizes in range, azimuth, and Doppler dimensions are adjusted in real-time according to target type, distance, and clutter levels, allowing the system to optimize clustering accuracy for different scenarios without requiring exhaustive computation for every possible condition.
Data Source
AI summary
Disclosed herein are systems and methods for adaptive clustering of points from radar scans, utilizing feedback information to define appropriate clustering gate sizes, and optimizing overall radar system performance in various scenarios and environments. Various approaches enable separation of over-clustered targets into multiple clusters by removing points whose distance from a weighted center exceeds a disparity threshold, and uniting under-clustered targets into a single cluster. Outlier points can be removed based on their SNR and Doppler relative to a median SNR and median Doppler of all points in the cluster. The optimal cluster center is identified based on variable weighting techniques that consider at least one dimension beyond spatial dimensions, such as Doppler information, signal-to-noise ratio information, environmental conditions, or system constraints. Adaptive gate sizing is performed to fine-tune clustering parameters based on feedback from various sources, such as target characteristics, environmental conditions, or system constraints, using previous scan data.