Parking Slot Corner Correction for Stable Rectangular Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current parking lot detection algorithms suffer from jitter in video streams, leading to irregular quadrangles and unpredictable deviations in automatic parking due to deviations in corner point detection between adjacent frames, especially in vertical, horizontal, and small-angle inclined parking lots.
Innovation Solution
A method and system for correcting parking lot corner points by performing rectangular correction based on the position relationship between detected corner points, using confidence levels and offsets to refine corner point accuracy, and employing geometric relationships to ensure accurate rectangularization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If corner point detection is performed in video streams, then parking lot detection can be achieved, but jitter occurs due to deviations between adjacent frames causing irregular quadrangles
Solution Approach 1:
The patent applies preliminary action by performing rectangular correction on detected corner points before using them for parking lot formation. The system proactively corrects potential deviations by enforcing rectangular constraints on the four corner points, preventing jitter from propagating into irregular quadrangles. This preliminary correction ensures that even if individual corner points have detection errors, the overall parking lot shape maintains geometric consistency.
2Ease of operation
If corner point detection allows certain deviation range, then detection results are considered correct without jitter in single-frame, but deviations between adjacent frames cause jitter in video streams
Solution Approach 1:
The patent implements feedback by using the detected corner points to form a parking lot shape, then evaluating whether this shape meets rectangular constraints. If the formed quadrangle deviates from rectangular properties, the system feeds back correction information to adjust the corner points. This closed-loop feedback mechanism ensures that detection results maintain both tolerance for individual point variations and consistency across multiple frames by enforcing global geometric constraints.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The invention relates to a parking lot corner point correction method and system. The method includes: an obtaining step of obtaining a parking lot corner point of a parking lot; and a correction step of performing, based on a position relationship between parking lot corner points of the parking lot, rectangular correction on the parking lot corner points to obtain a corrected parking lot. The obtaining step includes: a first obtaining sub-step of obtaining a parking lot corner point of a parking lot; a determination sub-step of determining whether a corner point confidence level of the parking lot corner point meets a first threshold and whether a corner point offset thereof meets a second threshold, and jumping to the correction step if it is determined that the corner point confidence level meets the first threshold and the corner point offset meets the second threshold, otherwise, proceeding to a second obtaining sub-step; and the second obtaining sub-step of discarding the parking lot corner point of the parking lot that is detected in the first obtaining sub-step, and then using a parking lot corner point at a corresponding position that is predicted by an odometer, as the parking lot corner point of the parking lot. According to the invention, the parking accuracy may be improved.