3D Radar Occupancy Grid Mapping with CNN-Based Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle assistance and control systems, such as navigation and driver assistance systems, face challenges in accurately detecting and understanding the vehicle's external environment, particularly in generating effective occupancy maps for radar-based systems.
Innovation Solution
A computer-implemented method for driving assistance in vehicles, which involves generating a three-dimensional occupancy grid map (3D OGM) and feature grid maps (FGMs) from radar point sensor data. These maps are then refined using a convolutional neural network (CNN), incorporating techniques like two-dimensional convolutions, pyramid processing, and adaptive re-centering to improve the accuracy and relevance of the maps for vehicle assistance systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If radar point sensor data is used to generate occupancy grid maps, then the system can detect the vehicle's external environment, but the accuracy and quality of the generated maps are insufficient for effective navigation and driver assistance
Solution Approach 1:
The patent transforms 2D radar point sensor data into a 3D occupancy grid map representation, adding a vertical dimension to the environmental model. This dimensional transformation enables more accurate representation of objects and obstacles in three-dimensional space, directly improving measurement precision while maintaining detection reliability through the enhanced spatial modeling capability
Solution Approach 2:
The system performs preliminary processing of radar data by generating feature grid maps that extract and organize key environmental features before final occupancy map generation. This preliminary action of feature extraction and organization improves the quality and accuracy of the final occupancy map by preparing refined input data structures
2Loss of information
If multiple feature grid maps are generated from radar data, then the representation of environmental features is enhanced, but the processing complexity and computational requirements increase
Solution Approach 1:
The patent segments the environmental data processing into multiple specialized feature grid maps, each capturing specific aspects of the environment (e.g., static objects, moving objects, terrain features). This segmentation preserves comprehensive environmental information by distributing it across multiple focused representations, while the modular structure actually reduces processing complexity compared to handling all features in a single undifferentiated data structure
Solution Approach 2:
The system merges multiple feature grid maps into a unified refined occupancy grid map that integrates all environmental features. This merging process consolidates the information from multiple sources into a single comprehensive representation, reducing the complexity of managing separate data structures while preserving all essential environmental information
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Computerized methods and system for driving assistance in a vehicle are provided. Based on radar point sensor data of the vehicle's environment, a three-dimensional occupancy grid map and a number of feature grid maps with a respective feature dimension corresponding to a feature of the radar point sensor data obtained by a radar system of the vehicle are generated which are then used to generate the refined occupancy grid map. The refined OGM may be provided to an assistance system of the vehicle.