Parking Space Detection Using Center and Corner Probability Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional parking space detection methods are complex and slow, and they cannot detect multiple types of parking spaces in a single detection process.
Innovation Solution
A method and apparatus that utilize a learning process on a top-view image to generate a center-point probability map and four inner-angular-point probability maps, enabling rapid detection of parking spaces through a single stage, independent of the type of parking space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple stages are used to detect parking spaces (detecting intersection points, determining parking space information, estimating additional points), then detection accuracy can be maintained, but detection speed decreases and process complexity increases
Solution Approach 1:
The conventional multi-stage detection process is segmented into three separate tasks: detecting intersection points of lines, determining parking space information based on intersection pairs, and estimating additional points. This segmentation allows each stage to be optimized independently and enables parallel processing possibilities, resolving the contradiction between maintaining accuracy through multiple stages and improving detection speed.
Solution Approach 2:
The patent performs preliminary detection of intersection points and parking space information in advance, storing these results for subsequent use. By pre-computing these intermediate results, the system avoids redundant calculations during actual parking space detection, thereby improving detection speed while maintaining accuracy through the use of pre-established intersection point data.
2Productivity
If a single detection process is used to detect multiple types of parking spaces, then detection efficiency improves, but the complexity of handling different parking space types increases
Solution Approach 1:
The patent creates a universal detection framework that can handle multiple types of parking spaces (perpendicular, parallel, angled, and curved parking spaces) through the same intersection point detection and probability map generation process. By using a unified approach with type-specific probability thresholds and mapping rules, the system achieves multi-functionality without significantly increasing process complexity, as the core detection algorithm remains the same across different parking space types.
Data Source
AI summary
A method and an apparatus for detecting a parking space, a device and a computer-readable storage medium are provided. The method includes: acquiring a top-view image of a scene; performing a learning process on the top-view image to obtain a center-point probability map and four inner-angular-point probability maps, and the center-point probability map includes a predicted probability of each pixel in the top-view image being a center point, each of the inner-angular-point probability maps includes a predicted probability of each pixel in the top-view image being an inner angular point, and the four inner-angular-point probability maps correspond to different types of inner angular points of a parking space; and obtaining a parking space detection result for the top-view image based on the center-point probability map and four inner-angular-point probability maps.