Containerized Sensor System for Real-Time Parking Space Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning systems face challenges in providing real-time responses due to intensive computation requirements for model training, making it difficult to manage tasks efficiently in applications like automobile parking space detection.
Innovation Solution
A system utilizing containerized sensors, a machine learning center, and a cloud-based network to monitor and record surrounding information, generate labeled data, and update models through a virtuous cycle, enabling real-time data processing and model refinement for improved parking space detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional machine learning systems perform intensive model training computation, then model accuracy is improved, but real-time response capability deteriorates
Solution Approach 1:
The system segments the machine learning workflow into distinct phases: data collection by onboard sensors, preliminary processing and labeling, cloud-based model training, and deployment of trained models to vehicles. This segmentation allows intensive computation to occur in the cloud while maintaining real-time response capability at the edge devices.
Solution Approach 2:
The patent introduces cloud-based infrastructure and automated labeling services as intermediaries between data collection and model training. These intermediaries handle the intensive computation burden, allowing vehicle-based systems to maintain real-time response capability without sacrificing model accuracy.
2Reliability
If more data is collected and processed for model training, then model performance is improved, but computational complexity and processing time increase
Solution Approach 1:
The system performs preliminary actions by collecting and labeling data in advance during normal vehicle operation. Labeled data is accumulated and prepared before model training begins, reducing the computational complexity and processing time required during actual training cycles.
Solution Approach 2:
The system implements self-service through automated labeling processes that use existing sensor data and algorithms to label training data without extensive human intervention. This reduces computational complexity while maintaining model performance through continuous learning from self-labeled data.
3Speed
If real-time parking space detection is implemented in vehicles, then responsiveness is improved, but onboard computational requirements and energy consumption increase
Solution Approach 1:
The detection system is segmented into lightweight onboard components for real-time inference and heavy cloud-based components for model training. The onboard sensors and processors handle only the essential real-time detection tasks, minimizing energy consumption while maintaining responsiveness.
Solution Approach 2:
The patent replaces heavy onboard mechanical computation with cloud-based computational resources. Trained models are deployed to vehicles for efficient inference, while model updates and retraining occur in the cloud, reducing the computational burden and energy consumption of onboard systems.
Data Source
AI summary
A method or system capable of managing automobile parking space (“APS”) using containerized sensors, machine learning center, and cloud based network is disclosed. A process, in one aspect, monitors the surrounding information observed by a set of onboard sensors of a vehicle as the vehicle is in motion. After selectively recording the surrounding information in accordance with instructions from a containerized APS model which is received from a machine learning center, an APS and APS surrounding information are detected when the vehicle is in a parked condition. Upon rewinding recorded surrounding information leading up to the detection of APS, labeled data associated with APS is generated based on APS and the recorded surrounding information. The process subsequently uploads the labeled data to the cloud based network for facilitating APS model training at the machine learning center via a virtuous cycle.