Roadside Image Processing Architecture for Low-Latency Safety Hints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The deployment of edge computing nodes near roadside devices in vehicle networks increases operational costs and requires significant computing resources due to the need for high-performance computing devices and frequent model retraining to adapt to location-specific data, leading to inefficiencies and high bandwidth consumption.
Innovation Solution
A three-layer architecture is implemented, where a first computing node near the roadside device intercepts sub-images for object detection, a second computing node performs classification and risk assessment, and a cloud computing node handles model training and deployment, reducing the need for high-power computing at the edge and enabling efficient model updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If computing power is deployed close to roadside devices, then latency is reduced and bandwidth is improved, but operational cost significantly increases
Solution Approach 1:
The computing system is segmented into three distinct layers: roadside devices for data collection, edge computing nodes for real-time processing, and cloud computing centers for model training. This segmentation allows low-latency processing at the edge while avoiding the high cost of deploying full computing power at every roadside device.
Solution Approach 2:
The system transitions from a two-layer architecture (roadside device directly to cloud) to a three-layer architecture by introducing an intermediate edge computing node layer. This dimensional change enables real-time processing close to roadside devices without requiring expensive high-performance computing hardware at each roadside device location.
2Ease of manufacture
If all computing power is deployed on far edge side, then operational cost is reduced, but latency increases and bandwidth consumption increases
Solution Approach 1:
Computing tasks are segmented and distributed across three layers: lightweight real-time processing at the edge node, model training at the cloud layer, and data collection at roadside devices. This segmentation allows cost-effective deployment while maintaining low latency for critical processing.
Solution Approach 2:
An intermediate edge computing node is introduced between roadside devices and the cloud computing center. This intermediary handles real-time processing locally to reduce latency, while only necessary data is transmitted to the cloud for model training, optimizing both cost and performance.
3Adaptability or versatility
If models are frequently retrained to adapt to location-specific data, then adaptability is improved, but computational resource consumption increases
Solution Approach 1:
The system adds a cloud computing center layer for model training, separating the adaptation function from the execution function. Location-specific models are trained at the cloud layer using historical data, then deployed to edge nodes for execution, reducing the computational burden on edge devices while maintaining location-specific adaptability.
Solution Approach 2:
Models are pre-trained at the cloud computing center using historical data from multiple locations before being deployed to edge nodes. This preliminary training action allows edge nodes to perform real-time processing without needing to retrain models locally, reducing computational resource consumption while maintaining adaptability to location-specific conditions.
Data Source
AI summary
Embodiments of the present disclosure provide an image processing method, an electronic device, and a computer program product. The method may include receiving a sub-image containing a target object and location information of the target object from a first computing node, the sub-image being intercepted from a monitoring image by the first computing node through target detection, the monitoring image being acquired by a roadside device and containing the target object, and the first computing node being located near the roadside device. The method may further include determining classification information of the target object based on the sub-image at a second computing node different from the first computing node. In addition, the method may further include generating safety hint information for the target object at the second computing node at least based on the classification information and the location information.