Distributed CNN Sensor Fusion for Lower Bandwidth Edge Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current early fusion networks in autonomous driving systems require significant memory and processing resources, are inefficient in data transmission, and rely on centralized general-purpose processors, which can be costly and power-intensive.
Innovation Solution
Implementing a distributed convolutional neural network architecture that preprocesses raw sensor data at edge devices, using specialized processors for feature extraction and downsampling, and transmitting reduced data to a central processor for further processing, thereby reducing network load and enhancing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single centralized neural network processes all sensor data, then detection accuracy is improved, but memory requirements and processing throughput increase significantly
Solution Approach 1:
The patent divides the centralized neural network into distributed neural networks deployed across multiple edge devices. Each edge device processes local sensor data independently, segmenting the overall processing task. This segmentation reduces the memory burden on any single device while maintaining detection accuracy through coordinated processing across the distributed network.
2Measurement precision
If a single centralized neural network processes all sensor data, then detection accuracy is improved, but processing throughput requirements increase
Solution Approach 1:
The patent segments the centralized processing architecture into multiple distributed edge devices, each handling a portion of the sensor data stream. This parallelization of processing tasks across multiple devices reduces the throughput burden on any single processor while collectively maintaining the detection accuracy of the original centralized system.
3Measurement precision
If raw sensor data is transmitted to a centralized processor, then comprehensive analysis is achieved, but network bandwidth consumption increases
Solution Approach 1:
The patent applies preliminary processing actions at the edge devices before data transmission. Each edge device performs initial sensor data processing, filtering, and feature extraction locally, then transmits only the processed results to the centralized system. This preliminary action reduces the volume of data requiring network transmission while preserving the comprehensive analysis capability through subsequent centralized integration of the processed data.
4Adaptability or versatility
If general-purpose processors are used for sensor fusion, then system flexibility is maintained, but processing efficiency decreases
Solution Approach 1:
The patent implements local quality optimization by deploying specialized neural network processing units at edge devices. Each edge device can be optimized for specific sensor types or processing tasks, improving local processing efficiency. The distributed architecture maintains overall system flexibility through the ability to independently configure and optimize each edge node for different requirements.
5Loss of energy
If all processing is performed at edge devices, then network load is reduced, but centralized coordination capability is lost
Solution Approach 1:
The patent segments the processing architecture into distributed edge devices that perform local processing to reduce network load, while maintaining a centralized coordination layer that receives processed data from edges and performs global decision-making. This segmentation enables both local autonomy at edge devices and centralized coordination for tasks requiring global context, optimizing the balance between network efficiency and automated decision-making capability.
Data Source
AI summary
An early fusion network is provided that reduces network load and enables easier design of specialized ASIC edge processors through performing a portion of convolutional neural network layers at distributed edge and data-network processors prior to transmitting data to a centralized processor for fully-connected/deconvolutional neural networking processing. Embodiments can provide convolution and downsampling layer processing in association with the digital signal processors associated with edge sensors. Once the raw data is reduced to smaller feature maps through the convolution-downsampling process, this reduced data is transmitted to a central processor for further processing such as regression, classification, and segmentation, along with feature combination of the data from the sensors. In some embodiments, feature combination can be distributed to gateway or switch nodes closer to the edge sensors, thereby further reducing the data transferred to the central node and reducing the amount of computation performed there.