Multi-Lane Road Geometry Estimation Using Bayesian Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current road estimation systems for autonomous vehicles face challenges in accurately modeling complex road geometries, particularly in scenarios with multiple lanes, on-ramps, and off-ramps, due to limitations in data association and clutter handling, which affects the reliability and redundancy of road feature detection.
Innovation Solution
A model-based algorithm using a Bayesian approach with a multiple hypothesis tracker, employing a filter state that predicts road model hypotheses based on vehicle motion data and perception data from sensors, associating road features to determine lane existence probabilities and selecting the most probable road model for accurate road geometry estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a traditional road estimation system is used, then the system complexity is low, but the accuracy of road geometry estimation deteriorates in complex environments with multiple lanes, on-ramps, and off-ramps
Solution Approach 1:
The road estimation problem is segmented into multiple hypotheses, each representing a different road configuration (e.g., different numbers of lanes, on-ramp positions). The system maintains and evaluates multiple road model hypotheses simultaneously, allowing it to handle complex road geometries by breaking down the estimation task into manageable segments rather than attempting to solve the entire problem at once.
Solution Approach 2:
The system dynamically changes parameters of road model hypotheses based on sensor data. Each hypothesis contains parameters such as lane positions, curvatures, and existence probabilities. As new perception data arrives, these parameters are updated to reflect the most likely road configuration, enabling accurate adaptation to complex and changing road geometries.
2Reliability
If multiple road model hypotheses are maintained to improve estimation accuracy, then the reliability of road feature detection improves, but the computational redundancy increases
Solution Approach 1:
The system dynamically manages multiple road model hypotheses by continuously updating their existence probabilities based on sensor data. Hypotheses that become increasingly unlikely are pruned or merged, while promising hypotheses are maintained and refined. This dynamic approach ensures high detection reliability through multiple hypotheses while controlling computational redundancy by adaptively managing the hypothesis set size.
Solution Approach 2:
The system employs feedback mechanisms where perception data from sensors is continuously compared against predictions from multiple road model hypotheses. The existence probabilities of hypotheses are updated based on how well they explain the observed road features. This feedback loop ensures that only hypotheses consistent with actual sensor data are maintained, improving reliability while eliminating computationally redundant hypotheses that contradict observations.
3Adaptability or versatility
If real-time updating of road model is implemented, then the adaptability to complex environments improves, but the processing time increases
Solution Approach 1:
The system performs preliminary actions by pre-defining a set of plausible road model hypotheses with typical parameters and structures. These hypotheses are prepared in advance with reasonable assumptions about road geometries (e.g., standard lane widths, common curvature ranges). When processing sensor data, the system only needs to evaluate and update these pre-prepared hypotheses rather than generating road models from scratch, significantly reducing processing time while maintaining adaptability to complex environments.
Data Source
AI summary
A computer-implemented method for estimating a road geometry of a road upon which a vehicle is traveling and related aspects are disclosed. The disclosed embodiments provide a model-based technique for estimating a road model having an arbitrary number of lanes on a road as well as any entrance lanes (i.e. on-ramp or slip roads) and exit lanes (i.e. off ramps or off-slip roads). In more detail, the herein disclosed embodiments utilize a Bayesian approach to multi-lane tracking in various traffic scenarios, such as e.g. highway scenarios. The employed model-based algorithm estimates the lane center curves of multi-lane roads based on vehicle motion data (e.g. speed, angular velocity) and perception data (e.g. camera output, lidar output, radar output, etc.) that includes detected road features (e.g. lane markers, road edges, road barriers, guard rails, road markers, etc.) or other objects (e.g. other road users).