Parking Space Detection Model Using Spatial Transformation and Logistic Regression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for detecting parking spaces in outdoor parking lots face challenges due to inter-occlusion, shadow, lighting, and perspective issues, as well as variations in pavement types and non-uniform parking space dimensions, which affect classification accuracy.
Innovation Solution
A method involving a computer system that transforms and extracts features from training images using spatial and logistic regression parameters to establish an optimal parking space detection model, adapting to image variations and improving classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single criteria is used for classification, then the classification process is simple, but the accuracy of classification deteriorates due to variations in pavement types and parking space dimensions
Solution Approach 1:
The patent segments the classification process into multiple independent stages: spatial transformation parameters are optimized separately from feature extraction parameters, which are in turn optimized separately from logistic regression parameters. Each segment addresses specific aspects of the classification problem (geometric normalization, texture/shape feature extraction, and final classification), allowing complex multi-type pavement and dimension variations to be handled through a series of specialized processing steps rather than a single monolithic classifier.
2Reliability
If camera geometry and projection matrix are utilized to select regions, then inter-occlusion and perspective problems are alleviated, but the method still fails to address variations in pavement types and dimensions
Solution Approach 1:
The patent applies local quality by using spatial transformation parameters that are specifically tailored to normalize the geometric characteristics of parking spaces in the image. These parameters adjust the spatial coordinates and dimensions of detected parking spaces to account for perspective distortion and occlusion, ensuring that each parking space is properly oriented and scaled regardless of its position in the image or the camera angle.
Solution Approach 2:
The patent employs parameter changes by optimizing spatial transformation parameters, feature extraction parameters, and logistic regression parameters to adapt the classification system to different pavement types and parking space dimensions. The spatial transformation parameters adjust geometric properties, while feature extraction parameters are tuned to capture texture and shape characteristics specific to different pavement materials, enabling the system to handle diverse parking lot environments.
3Productivity
If conventional classification methods are used, then the processing is fast, but accuracy deteriorates due to shadow and lighting problems
Solution Approach 1:
The patent applies preliminary action by performing spatial transformation and feature extraction before the final classification step. The spatial transformation parameters pre-process the image data to normalize parking space geometries and reduce the impact of shadows and lighting variations. Feature extraction parameters then extract robust texture and shape features that are less sensitive to lighting conditions, preparing the data in advance for more accurate classification without requiring complex real-time adjustments during the classification itself.
Data Source
AI summary
A method to be implemented by a computer system includes: transforming each of training images into a respective transformed training image; calculating a classification loss value based on detection results that are acquired from parking status prediction results obtained by performing feature extraction on the transformed training images; and adjusting candidate spatial transforming parameters, candidate feature extraction parameters and candidate logistic regression parameters when it is determined that a combination of the aforementioned parameters is not optimal based on the classification loss value, followed by repeating above-mentioned steps using the adjusted parameters.