Edge-Cloud Object Detection Using Single-Feature Transmission
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Edge devices face challenges in performing real-time object detection due to resource constraints and long image transmission times when using complex deep neural networks, especially in object detection tasks that require multiple feature layers.
Innovation Solution
A cloud-server-edge collaboration method where an edge device extracts a single feature from a predetermined layer and transmits it to a cloud server, which uses a reconstructor to generate intermediate features for object detection, optimizing resource usage and reducing transmission time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If edge devices transmit images to cloud server for processing, then detection accuracy is improved, but transmission time increases
Solution Approach 1:
The patent extracts only the necessary feature map from the edge device instead of transmitting the entire image. By identifying and transmitting only the relevant feature data required for object detection, the system reduces transmission time while maintaining detection accuracy, as the cloud server receives pre-processed essential information rather than raw image data.
Solution Approach 2:
The patent segments the object detection task between edge device and cloud server. The edge device performs initial feature extraction and transmits only the feature map, while the cloud server completes the detection process. This segmentation allows the system to leverage both edge computing speed and cloud computing power, reducing overall transmission time while maintaining accuracy.
2Productivity
If edge devices perform real-time inference with complex deep neural networks, then detection speed is improved, but resource consumption increases
Solution Approach 1:
The patent segments the computational workload by dividing the deep neural network inference between edge device and cloud server. The edge device performs feature extraction using a portion of the network, then transmits the feature map to the cloud server for completion of the detection task. This segmentation enables real-time processing at the edge while offloading computationally intensive operations to the cloud, balancing detection speed and resource consumption.
3Measurement precision
If multiple feature layers are transmitted for object detection, then detection accuracy is improved, but transmission amount increases
Solution Approach 1:
The patent extracts only the essential feature map from multiple available feature layers at the edge device. Instead of transmitting all feature layers which would contain redundant information, the system identifies and transmits only the most relevant feature representation. This extraction approach maintains detection accuracy by preserving critical object information while significantly reducing the transmission amount.
Data Source
AI summary
The present disclosure relates to a deep neural network-based object detection device, system, and method, and more particularly, to a deep neural network-based object detection device, system, and method capable of rapidly detecting an object in an image in real time.