Lane Boundary Tracking Using Past Partition Line Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems struggle to accurately recognize lane partition lines due to sensor instability, often leading to the misidentification of multiple partition lines, which can hinder precise lane recognition for automated driving.
Innovation Solution
A mobile object control device that utilizes a partition line candidate recognizer to identify potential lane partition lines from captured images and a partition line searcher to select the closest line to a past partition line, based on distance and alignment, to accurately determine the current lane boundaries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sensor-based lane detection is used to recognize partition lines, then automated driving control can be implemented, but recognition accuracy deteriorates when partition lines are lost or interrupted, leading to misidentification of multiple partition lines
Solution Approach 1:
The system performs preliminary action by predicting the expected position of partition lines based on historical detection data and road geometry before actual detection occurs. This allows the system to have a reference position ready in advance, improving reliability when detection is difficult.
Solution Approach 2:
The system implements feedback by continuously comparing detected partition line positions with predicted positions and using this information to adjust subsequent detections. The loss function incorporates the difference between predicted and actual positions, creating a feedback loop that improves measurement precision over time.
2Adaptability or versatility
If multiple partition line candidates are recognized from sensor data, then comprehensive lane coverage is achieved, but selection accuracy deteriorates due to difficulty in determining which candidate is the actual partition line
Solution Approach 1:
The system uses feedback by calculating a loss function that measures the difference between predicted partition line positions and actual detected candidates. This feedback mechanism enables accurate selection among multiple candidates by identifying which one minimizes the position difference.
Solution Approach 2:
The system applies parameter changes by transforming the selection problem into an optimization problem where the loss function value serves as the selection criterion. By changing the selection parameter from arbitrary choice to loss-minimization, the system achieves precise identification among multiple candidates.
3Speed
If partition line detection relies solely on current sensor data, then real-time responsiveness is maintained, but accuracy deteriorates due to instability of sensor recognition
Solution Approach 1:
The system performs preliminary action by predicting partition line positions using historical data and road geometry before current detection occurs. This allows the system to maintain real-time responsiveness while improving accuracy through pre-computed reference positions.
Solution Approach 2:
The system applies beforehand cushioning by using predicted partition line positions as a reference buffer against sensor instability. When sensor detection is unreliable, the predicted position serves as a cushion that prevents large errors, maintaining both speed and precision.
Data Source
AI summary
A mobile object control device of an embodiment includes a partition line candidate recognizer that recognizes candidates for partition lines for partitioning a traveling lane in which a mobile object travels from a first image including the vicinity of the mobile object captured by an image capturer, and a partition line searcher that selects a partition line candidate closest to a past partition line for partitioning the traveling lane of the mobile object before a predetermined time, among the partition line candidates recognized by the partition line candidate recognizer, as a partition line for partitioning a current traveling lane of the mobile object.