Road Shape Estimation Using Point Cloud and Position Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous driving technologies face challenges in accurately estimating road shapes, which can lead to decreased safety and reliability in autonomous control operations.
Innovation Solution
A road shape estimation method that uses an estimated-point cloud acquired by object recognition sensors mounted on a vehicle, compensating for accuracy decreases through learning based on complex position data of the vehicle, including three-dimensional position and attitude data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If road shape estimation is performed using estimated-point cloud from object recognition sensors, then autonomous driving control can be enabled, but accuracy of road shape estimation decreases
Solution Approach 1:
The patent applies feedback by continuously comparing the estimated road shape from point cloud data with actual road conditions detected by sensors, and adjusting the estimation based on the discrepancies. This feedback mechanism enables the system to maintain high accuracy in autonomous driving control despite the inherent limitations of point cloud-based estimation methods.
Solution Approach 2:
The patent changes parameters by transforming the road shape estimation from a static geometric model to a dynamic model that incorporates temporal variations and environmental factors. By adjusting estimation parameters based on real-time sensor data and learned patterns, the system achieves both automation and high measurement precision simultaneously.
2Measurement precision
If learning-based compensation is applied to improve road shape estimation accuracy, then measurement precision improves, but device complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-training neural networks and establishing compensation models before actual road shape estimation occurs. The system learns from extensive training data and pre-computes correction factors, which are then applied during real-time operation. This preliminary preparation reduces the computational burden during actual estimation while maintaining high accuracy, thus managing device complexity effectively.
Solution Approach 2:
The patent introduces an intermediary compensation model that mediates between the raw point cloud data and the final road shape estimation. This intermediary layer processes and refines the estimation data, separating the complex learning tasks from the real-time estimation process. The intermediary model handles complexity by pre-processing and filtering, allowing the main estimation system to remain relatively simple while achieving high precision through the learned compensation relationships.
Data Source
AI summary
A processor of a road shape estimation apparatus is configured to execute road-shape estimation program instructions to accordingly (i) estimate a shape of a road located on a traveling course of an own vehicle based on an estimated-point cloud acquired by at least one object recognition sensor mounted to the own vehicle, the estimated-point cloud comprising an assembly of estimated points on the road located on the traveling course of an own vehicle, and (ii) compensate for a decrease in an estimation accuracy of the shape of the road based on the estimated-point cloud using a result of learning, based on complex position data of the own vehicle, information about the shape of the road, the complex position data of the own vehicle including at least one of three-dimensional position of the own vehicle and attitude data of the own vehicle.