CNN Parking Space Detection Using Point Regression and Line Relations

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current technologies for autonomous parking lack efficient methods for detecting parking spaces, relying heavily on sensors and lacking comprehensive studies on using Deep Convolutional Neural Networks (CNNs) for this purpose, which could lead to financial losses due to inefficient parking practices.

Innovation Solution

A learning method and device utilizing a Convolutional Neural Network (CNN) to perform point regression and detect relationships between points in images, generating estimated probability distributions and relational linear segment information to identify unoccupied parking spaces, allowing for easier and more economical auto-parking systems.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If sensors are used to detect parking spaces, then detection reliability is improved, but device complexity and cost increase

Engineering Contradiction:
Improveparking space detection reliabilityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces traditional mechanical/optical sensors with a deep learning-based image processing system. A convolutional neural network (CNN) processes images from the vehicle's camera to detect parking spaces, replacing complex sensor arrays with a software-based solution that achieves comparable or superior detection reliability while reducing hardware complexity

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system creates a digital representation (segmentation map) of the parking environment from camera images. The CNN generates pixel-level labels that copy the physical parking space structure into a processed image format, enabling detection without direct sensor contact with the parking spaces

Inventive Principle:
Principle #26Copying

2Device complexity

If traditional image processing methods are used, then device complexity is reduced, but measurement precision and detection accuracy deteriorate

Engineering Contradiction:
Improvesystem complexityVSAvoidparking space detection accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent transforms the detection problem from traditional image processing parameters to deep learning parameters. The CNN architecture with multiple convolutional layers, activation functions, and pooling operations changes the processing parameters dynamically, enabling high-precision detection of parking space boundaries, angles, and occupancy status

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The system moves from 2D image processing to 3D feature extraction by incorporating spatial relationships, depth information from multiple image layers, and contextual features from the training data. The CNN processes images in multiple dimensions simultaneously, detecting not only parking space locations but also their geometric properties and relationships with surrounding objects

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Measurement precision

If deep learning models are trained with large datasets, then detection accuracy is improved, but loss of time during training increases

Engineering Contradiction:
Improveparking space detection accuracyVSAvoidtraining time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary actions by pre-training the CNN on a large dataset of labeled parking images before deployment. The model learns general parking space patterns, vehicle types, and environmental conditions in advance, so that during actual operation it can quickly adapt to new situations with minimal additional training time

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The training process is made dynamic through transfer learning and fine-tuning capabilities. The model can be initially trained on a large dataset for general accuracy, then quickly adapted to specific environments or conditions with smaller datasets, balancing training time investment with detection accuracy requirements for different deployment scenarios

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3690717B1Learning method and learning device, and testing method and testing device for detecting parking spaces by using point regression results and relationship between points to thereby provide an auto-parking system
Publication Date: 2023.12.13 STRADVISION
  • EP3690717B1 patent drawingFigure 1
  • EP3690717B1 patent drawingFigure 2
  • EP3690717B1 patent drawingFigure 3

AI summary

A learning method for detecting unoccupied parking spaces by using probability distributions on decision points of the unoccupied parking spaces and relational linear segment information on relationships among the decision points is provided. And the method includes steps of: (a) a learning device performing (i) a process of instructing a first CNN to apply a first CNN regression operation to a parking circumstance image, to thereby calculate each of one or more estimated probability distributions, and (ii) a process of instructing a second CNN to apply a second CNN regression operation to the parking circumstance image, to thereby generate estimated relational linear segment information; and (b) the learning device instructing a loss layer to perform (i) a process training parameters in the first CNN by performing backpropagation using a first loss, and (ii) a process of training of parameters in the second CNN by performing backpropagation using a second loss.