Radar Ground Truth Generation via Occupancy Grids
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating ground truth data for radar-based machine learning in driver assistance systems are limited by the need for expensive and time-consuming manual labeling or the use of additional sensors, which restricts the accuracy and efficiency of occupancy grid detection and underdrivability classification.
Innovation Solution
A computer-implemented method that generates ground truth data by utilizing sensor data from present, past, and future points in time, including the use of two maps (full-range and limited-range maps) to determine cell probabilities, allowing for automated labeling of underdrivable and non-underdrivable regions without requiring additional sensors or manual labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling or additional sensors (e.g., Lidar/camera) are used to generate ground truth data, then measurement precision and reliability are improved, but device complexity, cost, and time consumption increase
Solution Approach 1:
The radar sensor generates its own ground truth data by processing its own sensor data through occupancy grid determination and underdrivability classification algorithms, eliminating the need for external sensors or manual labeling
Solution Approach 2:
The method creates a virtual copy of the environment through occupancy grids and classification results that serve as ground truth data, replacing the need for physical additional sensors or manual creation of ground truth
2Measurement precision
If manual labeling or additional sensors are used to generate ground truth data, then measurement precision is improved, but productivity decreases
Solution Approach 1:
The radar system automatically generates ground truth data through self-processing of its sensor data using machine learning models, eliminating the time-consuming manual labeling process and enabling rapid automated ground truth generation
Solution Approach 2:
The patent replaces manual mechanical labeling processes with automated computational algorithms that process radar data to generate occupancy grids and underdrivability classifications, significantly improving generation speed
3Measurement precision
If additional sensors (e.g., Lidar/camera) are used to generate ground truth data, then measurement precision is improved, but cost increases
Solution Approach 1:
The radar sensor performs multiple functions: it both collects sensor data and generates ground truth data for machine learning training, eliminating the need for additional specialized sensors like Lidar or cameras
Solution Approach 2:
The existing radar sensor serves itself by processing its own data to create ground truth, removing the need to purchase and integrate additional expensive sensing hardware
4Measurement precision
If radar sensor data from future points in time is used to determine ground truth data, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The method collects and stores sensor data from multiple future time points before using them to determine ground truth data for earlier time points, allowing more complete information to be available for accurate labeling
Solution Approach 2:
The patent uses occupancy grids and underdrivability classification models as intermediaries that process and integrate information from multiple time points to produce accurate ground truth labels
Data Source
AI summary
A computer-implemented method for generating ground truth data may include the following steps carried out by computer hardware components: for a plurality of points in time, acquiring sensor data for a respective point in time; and for at least a subset of the plurality of points in time, determining ground truth data of the respective point in time based on the sensor data of at least one present and/or past point of time and at least one future point of time.