Radar Data Segmentation Using Space-Time Coordinates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing driver assistance systems face challenges in accurately segmenting radar data due to radar reflection sparsity, making it difficult to apply conventional point cloud processing deep learning techniques effectively.
Innovation Solution
A method and device that utilize a space-time representation by combining spatial and time-based coordinates, employing a trained artificial neural network to segment radar data, distinguishing between stationary and non-stationary objects and noise, using a first-in first-out memory to store past frames for enhanced reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional point cloud processing deep learning techniques are applied to radar data, then local neighborhood information can be captured precisely for every point, but the techniques cannot be effectively applied due to radar reflection sparsity
Solution Approach 1:
The patent transforms the sparse radar reflection data from spatial coordinates only to space-time coordinates by adding a time dimension. This dimensional extension allows the application of deep learning techniques that require sufficient data density, as the time dimension provides additional information about object dynamics and movement patterns, effectively resolving the sparsity problem while maintaining segmentation accuracy
Solution Approach 2:
The patent performs preliminary transformation of radar reflection coordinates from spatial to space-time representation before applying deep learning segmentation. By pre-processing the data to include time-based coordinates and calculating associated metrics (such as distance changes over time), the system prepares the data in a format suitable for conventional deep learning techniques, enabling effective segmentation despite initial sparsity
2Loss of information
If spatial coordinates only are used for radar reflections, then the data structure remains simple, but movement-dependent information and object dynamics cannot be captured
Solution Approach 1:
The patent adds a time dimension to the spatial coordinates of radar reflections, creating space-time coordinates. This additional dimension enables the capture of movement-dependent information and object dynamics without fundamentally complicating the coordinate system, as the time dimension can be integrated into existing deep learning architectures
Solution Approach 2:
The patent introduces time-based coordinates as an intermediary representation that bridges spatial position and object dynamics. These intermediate calculations (such as distance changes over time intervals) serve as mediators that encode movement information in a format that can be processed by standard deep learning segmentation algorithms
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enhances the accuracy of radar data segmentation, enabling reliable differentiation between different types of objects and noise, thereby improving the functionality of driver assistance systems.
Implementation Method 1
a radar sensor can emit radar signals which are reflected by objects in a surrounding of the vehicle. The coordinates of the reflecting objects can be determined by analyzing the received radar signals
Data Source
Figure 1~2
Figure 3
AI summary
A device for segmenting radar data of a vehicle comprises an interface which is configured to receive, from at least one radar sensor of the vehicle, radar data acquired at a plurality of measurement cycles. The device further comprises a computing device configured to identify, for each measurement cycle, radar reflections and to determine coordinates of the radar reflections, based on the radar data acquired at the measurement cycle, wherein the coordinates of each radar reflection comprise a time-based coordinate and spatial coordinates, wherein the spatial coordinates are defined relative to a global coordinate system. The computing device is further configured to provide the coordinates of the radar reflections of at least some of the measurement cycles as input to a trained artificial neural network. The computing device is further configured to receive a segmentation of the radar reflections as output of the trained artificial neural network.