Convolutional Neural Network Parking Detection via 3D Coordinates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional roadside parking management systems face inefficiencies and inaccuracies due to interference from environmental factors and the need for manual recording, leading to under-reporting and high labor costs, especially when detecting vehicles in complex or high-traffic conditions.
Innovation Solution
A parking detection method and device based on visual difference, utilizing real-time video frames from cameras, labeled with time information, and processed through a convolutional neural network to determine vehicle information, feature points, and three-dimensional coordinates, enabling accurate detection of vehicle status without reliance on conventional management devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If conventional roadside parking management devices (geomagnetic device, video pole, high-position dome camera, parallel matrix device) are used, then parking space management is initiated, but detection accuracy deteriorates due to interference from environmental factors and ground objects
Solution Approach 1:
The patent replaces conventional mechanical/detector-based parking management systems (geomagnetic devices, video poles, dome cameras, parallel matrix devices) with an image processing-based system using convolutional neural networks to detect vehicle features and calculate three-dimensional coordinates, thereby eliminating the detection inaccuracies caused by environmental interference in the original mechanical systems
Solution Approach 2:
The patent introduces an intermediary processing layer consisting of convolutional neural networks that process camera images to extract vehicle feature points and calculate three-dimensional coordinates. This intermediary processing step filters out environmental interference and provides accurate vehicle status detection, serving as a mediator between the camera and the parking management system
2Ease of manufacture
If geomagnetic device is used for low-cost parking detection, then cost is reduced, but detection accuracy deteriorates due to interference from ground objects
Solution Approach 1:
The patent replaces the geomagnetic device with an image processing system using convolutional neural networks. While the original geomagnetic device is low-cost, the new system achieves superior detection accuracy by processing visual information from cameras, extracting vehicle feature points, and calculating three-dimensional coordinates to determine vehicle entry and exit status
3Difficulty of detecting and measuring
If video pole is used for parking detection, then vehicle detection is enabled, but reliability deteriorates due to interference from people and low installation position
Solution Approach 1:
The patent replaces the video pole system with an image processing-based detection system using convolutional neural networks. The system processes camera images to extract vehicle feature points and calculate three-dimensional coordinates, providing reliable vehicle status detection that is not affected by the presence of people or installation position, thereby improving detection reliability
Solution Approach 2:
The patent transitions from two-dimensional image processing to three-dimensional spatial coordinate calculation. By calculating three-dimensional coordinates of vehicle feature points and determining vehicle status based on spatial position changes, the system achieves reliable vehicle detection that is not interfered with by people or camera installation position
4Ease of operation
If mobile vehicle patrolling is used for parking management, then vehicle supervision is enabled, but productivity deteriorates due to requirement for human participation and motor vehicle
Solution Approach 1:
The patent replaces mobile vehicle patrolling with an automated image processing system using convolutional neural networks. The system processes camera images to detect vehicle feature points and calculate three-dimensional coordinates, providing continuous automated vehicle status monitoring without requiring human participation or motor vehicles, thereby significantly improving parking management productivity and efficiency
5Difficulty of detecting and measuring
If high-position dome camera is used for parking detection, then vehicle monitoring is enabled, but measurement precision deteriorates due to hardware defects in identification when multiple vehicles enter or exit simultaneously
Solution Approach 1:
The patent replaces the high-position dome camera system with an image processing-based system using convolutional neural networks. The system processes camera images to extract vehicle feature points and calculate three-dimensional coordinates, providing accurate vehicle status detection even in heavy traffic conditions where multiple vehicles enter or exit simultaneously, thereby improving measurement precision
Solution Approach 2:
The patent segments the vehicle detection process into multiple stages: extracting vehicle feature points from images, calculating three-dimensional coordinates for each feature point, and determining vehicle status based on coordinate changes. This segmentation allows the system to accurately track individual vehicles even in heavy traffic conditions, improving vehicle identification precision
6Difficulty of detecting and measuring
If parallel matrix device is used for roadside management, then vehicle detection is enabled, but reliability deteriorates due to blockage by garden trees and branches
Solution Approach 1:
The patent replaces the parallel matrix device with an image processing system using convolutional neural networks. The system processes camera images to extract vehicle feature points and calculate three-dimensional coordinates, providing reliable vehicle status detection that is not affected by blockage from garden trees or branches, thereby improving detection reliability
Data Source
AI summary
A parking detection method based on visual difference includes: obtaining a video frame of a predetermined monitoring area captured by each camera in real time, and labeling the video frame corresponding to each camera with time information of a current moment; determining vehicle information of a to-be-detected vehicle in each video frame labeled with the time information through a predetermined convolutional neural network model; determining feature point information of the to-be-detected vehicle in each video frame according to the vehicle information of the to-be-detected vehicle in each video frame; calculating a position relationship between the to-be-detected vehicle in each video frame and respective corresponding camera, and constructing current three-dimensional coordinates of the to-be-detected vehicle according to the position relationship; and determining a parking status of the to-be-detected vehicle according to the current three-dimensional coordinates of the to-be-detected vehicle. A parking detection device based on visual difference is further provided.