Machine-Learned Radar Object Counting for Adaptive Angle Finding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Radar systems face challenges in accurately distinguishing between single-object and multiple-object scenarios in range-Doppler bins, leading to computational inefficiencies and inaccurate angle finding, which affects the performance of advanced driving assistance systems.
Innovation Solution
A radar system utilizing a machine-learned model to identify the number of objects in each range-Doppler bin by processing radar data to generate beam vectors and extracting features such as magnitude variation, signal-to-noise ratio, and subarray beam vector correlations, then selecting appropriate angle-finding techniques based on the identified number of objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional algorithms (FFT or super-resolution) are used to determine angle-finding techniques, then computational efficiency or accuracy can be achieved, but the classification accuracy of object number in range-Doppler bins remains low (around 60%), leading to computational inefficiency and angle-finding inaccuracy
Solution Approach 1:
The patent replaces traditional mechanical signal processing algorithms (FFT and super-resolution techniques) with a machine learning-based classification system. The machine learning model processes radar data to accurately identify the number of objects in each range-Doppler bin, achieving 96% classification accuracy. This substitution enables the system to automatically select appropriate angle-finding techniques based on object count, thereby improving both measurement precision and computational efficiency.
Solution Approach 2:
The patent changes the approach to object detection by introducing machine learning parameters and features (such as signal characteristics, spatial distribution, and temporal patterns) to classify the number of objects in range-Doppler bins. This parameter transformation enables more accurate distinction between single-object and multi-object scenarios, leading to better computational efficiency and angle-finding accuracy.
2Reliability
If low-accuracy classification techniques are used to determine object numbers, then the system remains simple, but angle-finding inaccuracy and increased response time occur, reducing ADAS feature efficacy
Solution Approach 1:
The patent substitutes traditional simple threshold-based classification with a machine learning-based system that processes multiple features (signal magnitude, spatial distribution, temporal characteristics) to accurately determine object numbers. This substitution improves angle-finding accuracy and reliability while the automated selection of angle-finding techniques based on object count maintains computational efficiency.
Solution Approach 2:
The machine learning model acts as an intermediary between radar data collection and angle-finding processing. It accurately classifies the number of objects in range-Doppler bins and mediates the selection of appropriate angle-finding techniques, thereby improving reliability and reducing false detections without significantly increasing overall system complexity.
3Measurement precision
If super-resolution techniques are used for multiple objects in range-Doppler bins, then angle-finding accuracy improves, but computational cost increases significantly compared to FFT
Solution Approach 1:
The patent implements a dynamic selection mechanism where the angle-finding technique is automatically adjusted based on the number of objects detected in each range-Doppler bin. When a single object is detected, the computationally efficient FFT algorithm is used. When multiple objects are detected, the more accurate but computationally intensive super-resolution technique is applied. This dynamic adaptation optimizes the balance between computational cost and angle-finding accuracy.
Solution Approach 2:
The machine learning-based object number classification system replaces manual or fixed threshold-based decision-making with an automated intelligent selection mechanism. This substitution enables the system to dynamically choose between FFT and super-resolution techniques based on detected object counts, optimizing computational resource allocation while maintaining high angle-finding accuracy.
Data Source
AI summary
This document describes techniques and systems related to a radar system using a machine-learned model to identify the number of objects within each range-Doppler bin. For example, the radar system includes a processor that obtains radar data associated with objects and processes the radar data to generate beam vectors. The processor then uses a machine-learned model to identify the number of objects within each range-Doppler bin using extracted features (e.g., magnitude variation, signal-to-noise ratio, subarray beam vector correlations) of the beam vectors. The processor selects a particular angle-finding technique based on whether a single object or multiple objects are identified. In this way, the described systems and techniques more accurately identify the number of object in each range-Doppler bin, thus improving the computational efficiency and robustness of subsequent angle finding.