Skewed Polygon CNNs for Perspective-Aware Parking Space Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional CNNs for parking space detection use axis-aligned rectangular anchor boxes, which fail to accurately represent non-rectangular parking spaces due to perspective projection, necessitating additional processing to delineate their bounds.
Innovation Solution
Employ a CNN to determine corner points of skewed polygons, such as quadrilaterals, with confidence values predicting entrance likelihoods and displacement values to accurately delineate parking spaces, using minimum aggregate distance for training positive samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If axis-aligned rectangular anchor boxes are used for parking space detection, then the CNN output is simple and straightforward, but additional processing is necessary to accurately identify the bounds of parking spaces due to perspective projection
Solution Approach 1:
The patent changes the geometric parameters of the anchor boxes from axis-aligned rectangles to skewed quadrilaterals that can rotate and skew to match the perspective projection of parking spaces. This allows the detection output to directly represent the actual shape and orientation of parking spaces in the image, eliminating the need for additional post-processing to determine accurate bounds.
Solution Approach 2:
The patent introduces additional degrees of freedom in the anchor box representation by allowing rotation and skewing parameters. Instead of only translating axis-aligned rectangles, the system now operates in a higher-dimensional parameter space that includes rotation angles and skew factors, enabling direct modeling of perspective-distorted parking spaces.
2Ease of manufacture
If conventional CNN with axis-aligned rectangles is used, then training with IoU is straightforward, but detection accuracy for non-rectangular parking spaces deteriorates
Solution Approach 1:
The patent extends the parameter space of anchor boxes to include rotation and skewing parameters, transforming them from simple 2D rectangles to 4-point quadrilaterals. This allows the model to directly learn the perspective-transformed shapes of parking spaces while maintaining a systematic training approach using generalized IoU calculations that work with arbitrary quadrilateral shapes.
3Measurement precision
If additional processing is applied to convert rectangular anchor boxes to accurate parking space bounds, then detection accuracy improves, but computational time increases
Solution Approach 1:
The patent performs the geometric transformation in advance by directly outputting skewed quadrilateral anchor boxes that already account for perspective projection. Instead of generating simple rectangles and then applying post-processing transformations, the system pre-computes the correct quadrilateral shapes during the detection phase, eliminating the need for subsequent conversion steps.
Data Source
AI summary
A neural network may be used to determine corner points of a skewed polygon (e.g., as displacement values to anchor box corner points) that accurately delineate a region in an image that defines a parking space. Further, the neural network may output confidence values predicting likelihoods that corner points of an anchor box correspond to an entrance to the parking spot. The confidence values may be used to select a subset of the corner points of the anchor box and/or skewed polygon in order to define the entrance to the parking spot. A minimum aggregate distance between corner points of a skewed polygon predicted using the CNN(s) and ground truth corner points of a parking spot may be used simplify a determination as to whether an anchor box should be used as a positive sample for training.