Autonomous Vehicle Perception System Using Segmented Neural Network Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in implementing high-performance deep neural networks due to constraints on computational resources and sensor costs, making it difficult to effectively perceive objects on roadways.
Innovation Solution
A perception system that processes images containing depth information to identify regions of interest, using a neural network trained to detect and classify objects, while also utilizing a training method that combines signals from high-quality reference sensors and lower-cost production sensors to infer depth information, allowing for efficient object detection and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-performance deep neural networks are used for object detection and classification, then detection accuracy and classification performance are improved, but computational resource requirements and sensor costs increase
Solution Approach 1:
The system segments the image processing task by first identifying regions of interest containing unspecified objects using traditional computer vision techniques, then applying neural networks only to these specific regions rather than the entire image. This segmentation approach maintains detection accuracy while significantly reducing computational resource requirements and power consumption in autonomous vehicles.
Solution Approach 2:
The system applies neural network processing partially - only to identified regions of interest rather than to the complete image. This partial action approach provides sufficient detection performance for safety-critical applications while avoiding the excessive computational burden of processing entire high-resolution images with deep neural networks.
2Measurement precision
If high-performance deep neural networks are used for object detection and classification, then detection accuracy and classification performance are improved, but power consumption increases
Solution Approach 1:
The system segments the image processing task by first identifying regions of interest containing unspecified objects using traditional computer vision techniques, then applying neural networks only to these specific regions rather than the entire image. This segmentation approach maintains detection accuracy while significantly reducing computational resource requirements and power consumption in autonomous vehicles.
Solution Approach 2:
The system applies neural network processing partially - only to identified regions of interest rather than to the complete image. This partial action approach provides sufficient detection performance for safety-critical applications while avoiding the excessive computational burden of processing entire high-resolution images with deep neural networks.
3Measurement precision
If high-quality sensors with accurate readings are used for perception, then measurement precision is improved, but device cost increases
Solution Approach 1:
The system uses traditional computer vision algorithms as an intermediary to pre-process images and identify regions of interest before applying neural networks. This intermediary step enables the use of lower-cost sensors by compensating for their limitations through algorithmic enhancement, achieving accurate object detection without requiring expensive high-performance sensors.
Solution Approach 2:
The system replaces reliance on expensive high-precision mechanical sensors with a combination of lower-cost sensors and computer vision algorithms. By substituting mechanical sensor precision with algorithmic image processing, the system achieves comparable detection accuracy while reducing sensor costs for commercial autonomous vehicle deployment.
Data Source
AI summary
A perception system, comprising a set of reference sensors; a set of test sensors and a computing device, which is configured for receiving first training signals from the set of reference sensors and receiving second training signals from the set of test sensors, the set of reference sensors and the set of test sensors simultaneously exposed to a common scene; processing the first training signals to obtain reference images containing reference depth information associated with the scene; and using the second training signals and the reference images to train a neural network for transforming subsequent test signals from the set of test sensors into test images containing inferred depth information.