Lane Course Determination Using Grid Map and Light-Dark Transitions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining a lane course for vehicles are unsatisfactory, particularly at low speeds or when the field of view is blocked, leading to inaccurate lane tracking and estimation of curvature or straight lines.
Innovation Solution
A method using an image acquisition unit to detect structures delimiting the driveable area, which are entered into a grid-based environment map, with continuous updates using odometric data and detection of light-dark and dark-light transitions to determine the lane course, even at low speeds and in obstructed areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a grid-based environment map with occupation probabilities is used to detect lane structures, then the system can identify lanes and boundaries in open areas, but the detection becomes unreliable when the field of view is blocked or when driving slowly due to insufficient measurements
Solution Approach 1:
The patent introduces an intermediary representation called 'lane hypotheses' that bridges the gap between sensor measurements and lane detection. These hypotheses are generated based on road geometry models and serve as a mediator that can fill in information when direct sensor measurements are unavailable due to occlusions or low speeds, thereby maintaining detection reliability without requiring continuous visual input
Solution Approach 2:
The system performs preliminary actions by pre-generating multiple lane hypothesis candidates based on expected road geometries before actual lane detection is needed. These pre-computed hypotheses are then validated against sensor data when available, but can be used directly when measurements are insufficient, allowing the system to maintain reliable lane tracking in obstructed areas without waiting for sufficient sensor input
2Duration of action of stationary object
If model-based filter algorithms like Kalman filter with clothoid model are used for tracking, then the system can maintain tracking when no measurements are available, but incorrect assumptions about lane angle or curvature cause the lane to 'rotate away' at low speeds
Solution Approach 1:
The patent applies dynamics by making the lane tracking system adaptable to different vehicle speeds. At higher speeds, the system relies more on model-based predictions for continuous tracking. At lower speeds, it increases reliance on actual sensor measurements and updates lane hypotheses more frequently, dynamically adjusting the balance between prediction and measurement to maintain both continuity and accuracy across varying operating conditions
Solution Approach 2:
The system implements feedback mechanisms where lane detection results are continuously validated against sensor measurements when available. The lane hypotheses are refined based on feedback from actual observed lane markings and boundaries, allowing the system to correct drift that would otherwise occur from incorrect initial assumptions about lane geometry, thereby maintaining measurement precision over extended tracking periods
3Measurement precision
If image processing algorithms detect structures based on light-dark transitions, then the system can identify lane markings with high contrast, but the detection fails when lane markings are obscured by vehicles or in areas with insufficient contrast
Solution Approach 1:
The patent creates a universal lane detection system that functions effectively across multiple conditions by combining multiple detection approaches. The system can detect high-contrast lane markings using traditional image processing, while simultaneously using lane hypotheses based on road geometry to detect or infer lane positions in low-contrast or obscured conditions, making the system versatile across varying visibility conditions without sacrificing precision in any single condition
Solution Approach 2:
Lane hypotheses serve as an intermediary that connects direct image processing results with inferred lane positions. When image processing successfully detects lane markings, the hypotheses are validated and refined. When markings are obscured or low-contrast, the hypotheses provide a bridge that maintains lane position estimation based on road geometry, allowing the system to adapt to various visibility conditions while maintaining detection precision through the intermediary hypothesis validation process
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method for determining the course of a lane for a vehicle (10), wherein structures which delimit a traffic-bearing space by means of at least one image capturing unit (2) and these structures are plotted in a map of the surroundings (20) which divides the surroundings of the vehicle into a two-dimensional grid structure (20a) of cells (21). According to the invention - by means of odometric data of the vehicle (10) the position in the grid structure (21) of the map of the surroundings (20) is determined and continuously updated, - the spacing and the direction of the vehicle (10) with respect to those cells (21b) of the grid structure (20a) of the map of the surroundings (20) which have structures delimiting the travel path and/or the lane is determined, - light-dark and dark-light transitions in the image data generated by the image capturing unit (2) are detected and plotted in the cells (21) of the grid structure (20a) of the map of the surroundings (20), and - the course of the lane (22) is determined from the cells with the detected light-dark and dark-light transitions.