Infrastructure Perception With Super-Resolution for Low-Latency ADAS
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in processing perception data for autonomous vehicles due to limited computing resources, leading to significant end-to-end latency and inability to meet latency requirements of advanced driver assistance systems (ADAS) applications.
Innovation Solution
An infrastructure-supported perception system that includes infrastructure perception sensors capturing reduced resolution and frame rate data, with servers executing super-resolution and motion prediction algorithms to enhance data resolution and frame rate, and prioritizing messages based on a relevance priority metric.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If perception data is captured at standard resolution and frame rate, then object detection accuracy is improved, but end-to-end latency increases and computing resources are overwhelmed
Solution Approach 1:
The system captures perception data at reduced resolution and frame rate (partial action) to decrease the amount of data processed, thereby reducing end-to-end latency while still meeting ADAS requirements through selective processing and super-resolution enhancement for critical objects
2Measurement precision
If perception data is captured at standard resolution and frame rate, then object detection accuracy is improved, but computing resource requirements increase beyond available infrastructure capacity
Solution Approach 1:
The system processes only a subset of perception data at full resolution (partial action) by capturing at reduced resolution and frame rate, then applies super-resolution enhancement selectively to objects of interest, thereby reducing overall computing resource requirements while maintaining detection accuracy for critical objects
Solution Approach 2:
The system changes the resolution and frame rate parameters of captured perception data from standard to reduced levels, then dynamically adjusts these parameters through super-resolution enhancement for specific objects, optimizing the balance between accuracy and computing resource usage
3Loss of time
If perception data is captured at reduced resolution and frame rate, then end-to-end latency is reduced, but object detection accuracy deteriorates
Solution Approach 1:
The system captures data at reduced resolution and frame rate parameters to decrease latency, then dynamically changes these parameters back to standard levels through super-resolution enhancement and motion prediction algorithms for objects of interest, achieving low latency while restoring detection accuracy
Solution Approach 2:
The system introduces super-resolution enhancement algorithms as an intermediary process between reduced-resolution capture and final object detection, restoring image quality and maintaining detection accuracy despite the intermediate reduced-resolution step that enables lower latency
Data Source
AI summary
An infrastructure-supported perception system for connected vehicle applications includes one or more infrastructure perception sensors that capture perception data having a reduced resolution and a reduced frame rate. The reduced resolution includes a reduced number of pixels for a given frame when compared to a standard resolution and the reduced frame rate captures data at a lower rate when compared to a standard frame rate. The infrastructure-supported perception system includes one or more controllers that are part of a connected vehicle. The controllers of the connected vehicle are in wireless communication with the one or more infrastructure perception sensors and one or more servers, and the one or more servers are in wireless communication with the one or more infrastructure perception sensors. The controllers receive dynamic information regarding one or more detected objects in an environment surrounding the connected vehicle from the one or more servers.