Dynamic Height Map Grid Spacing for Autonomous Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately recognizing objects at a distance due to the limitations of grid spacing in height maps, which affects the reliability of autonomous driving control, and increasing resolution is hindered by limited data throughput and real-time calculation requirements.
Innovation Solution
A method to dynamically adjust the grid spacing of height maps based on the vehicle's driving state, using sensors and image processing to generate depth maps, 3D point clouds, and standardize height information to detect dynamic obstacles, allowing for adaptive grid spacing adjustments to improve recognition accuracy and calculation speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If the spacing of the occupancy grid is increased, then the calculation speed is improved, but the resolution of disparity deteriorates and the accuracy of object recognition decreases
Solution Approach 1:
The patent applies different grid spacings to different regions of the height map based on their importance. Specifically, the grid spacing is adjusted according to the distance from the vehicle and the presence of objects, with smaller spacing (higher resolution) allocated to regions containing objects or closer to the vehicle, and larger spacing (lower resolution) allocated to distant regions without objects. This local differentiation resolves the contradiction by providing high resolution where needed for accurate object recognition while using lower resolution in other areas to maintain calculation speed.
2Measurement precision
If the spacing of the occupancy grid is decreased, then the accuracy of object recognition is improved, but the data throughput requirement increases and real-time calculation becomes impossible
Solution Approach 1:
The patent implements variable grid spacing where high-resolution grids (small spacing) are applied only to specific regions containing objects or critical areas, while low-resolution grids (large spacing) are applied to the rest of the scene. This selective application of different grid qualities ensures that object recognition accuracy is maintained in critical regions without requiring uniformly high resolution across the entire height map, thus managing data throughput requirements for real-time calculation.
3Measurement precision
If the resolution of disparity is increased indefinitely, then the accuracy of object recognition is improved, but the calculation processing is delayed and determination and control cannot be implemented in time
Solution Approach 1:
The patent reduces calculation processing time by applying high resolution (small grid spacing) only to local regions where objects are detected or where high accuracy is critical, rather than uniformly applying high resolution to the entire height map. This localized approach maintains object recognition accuracy in critical areas while significantly reducing the overall computational load and processing time, enabling real-time determination and control.
Solution Approach 2:
The patent dynamically adjusts the grid spacing based on the detected driving state and object positions. The system continuously monitors the environment and adapts the height map resolution in real-time, increasing resolution when objects are detected and maintaining lower resolution when the environment is clear. This dynamic adaptation ensures accurate object recognition when needed while minimizing calculation processing time during normal driving conditions.
Data Source
AI summary
A method of adjusting a grid spacing of a height map for autonomous driving, may include acquiring a 2D image of a region ahead of a vehicle, generating a depth map using depth information on an object present in the 2D image, converting the generated depth map into a 3D point cloud, generating the height map by mapping the 3D point cloud onto a grid having a predetermined size, and adjusting a grid spacing of the height map in consideration of the driving state of the vehicle relative to the object.