Bayesian Lane Assignment for ADAS Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current advanced driver-assistance systems (ADAS) face challenges in accurately and reliably associating traffic participants with traffic lanes due to uncertainties in sensor detections, leading to unreliable and robustness issues in tracking geometric assignments.
Innovation Solution
A computer-implemented method using a discrete probability distribution and Bayesian updating to predict the likelihood of traffic participants being on specific lanes, incorporating a Markov chain model and dynamic model that accounts for road conditions and sensor data from various sources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If geometric approaches are used to assign traffic participants to traffic lanes, then the assignment can be computed, but the reliability and robustness of tracking is poor due to uncertainty in sensor detections
Solution Approach 1:
The patent transforms the lane assignment problem from a deterministic geometric computation to a probabilistic parameter estimation problem. By representing lane occupancy as a discrete probability distribution over traffic lanes and using Bayesian updating with likelihood functions, the system incorporates uncertainty quantification while maintaining computability through iterative probability refinement.
Solution Approach 2:
The patent introduces a probabilistic model as an intermediary between sensor detections and lane assignment. The discrete probability distribution acts as a mediator that accumulates evidence from multiple sensor measurements over time, allowing the system to compute reliable lane assignments even when individual sensor detections are uncertain.
2Ease of manufacture
If geometric approaches are used to assign traffic participants to traffic lanes, then the assignment can be computed, but the robustness of tracking is poor
Solution Approach 1:
The patent implements continuous tracking by updating the discrete probability distribution at each time cycle using Bayesian updating. This continuous refinement of probability distributions over time creates a stable tracking system that maintains robustness through persistent probabilistic inference rather than discrete geometric re-computation.
Solution Approach 2:
The patent incorporates feedback mechanisms where the predicted discrete probability distribution from the dynamic model is continuously updated using likelihood functions based on new sensor measurements. This feedback loop allows the system to correct tracking errors and maintain robustness by continuously incorporating new evidence while preserving temporal consistency.
3Measurement precision
If sensor detections are used to determine traffic participant position, then position data is obtained, but uncertainty in the data leads to unreliable confidence in lane assignment
Solution Approach 1:
The patent transforms the static lane assignment problem into a dynamic probabilistic inference process. The discrete probability distribution evolves over time according to a dynamic model, allowing the system to adapt to changing sensor measurements and maintain reliable confidence estimates even when individual position measurements are uncertain.
Solution Approach 2:
The patent performs preliminary probabilistic inference by providing an initial discrete probability distribution before final lane assignment is made. This preliminary estimation allows the system to prepare confidence measures in advance and refine them through Bayesian updating, ensuring reliable confidence quantification is available before critical decisions are made.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
The invention provides a computer-implemented method for locating a traffic participant on a traffic lane of a plurality of traffic lanes. A discrete probability distribution is provided which indicates for each traffic lane a probability that the traffic participant is located on said traffic lane. Further, a likelihood function is provided which comprises for each traffic lane a probability distribution which indicates the probability for each lateral position that said lateral position is measured, under the assumption that the traffic participant is located on said traffic lane. The discrete probability distribution at a next cycle is predicted, based on the discrete probability distribution at a current cycle and using a dynamic model of the traffic participant. A measured lateral position of the traffic participant is determined, based on measurement data from at least one sensor. The predicted discrete probability distribution is updated by Bayesian updating, using the likelihood function and the measured lateral position.