Graph Convolution Driver Attention Reasoning for Key Object Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated driving in interactive scenarios faces challenges due to the need for scene analysis, situation understanding, intention prediction, decision-making, and planning, particularly with human eye gaze skewing determinations and requiring high computational processing for object detection and tracking in each frame, which is time-consuming and resource-intensive.
Innovation Solution
A computer-implemented method using a graph convolution network to analyze images of a vehicle's surroundings, determining relevant objects' importance scores to autonomously control the vehicle by identifying top relevant objects that influence its operation, thereby enhancing attention reasoning and reducing computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object detection and tracking is performed for each frame of a clip, then object-level reasoning accuracy is improved, but computational processing time increases
Solution Approach 1:
The system performs object detection and tracking only on selected key frames rather than every frame, preparing results in advance for reasoning tasks. This preliminary action on a subset of frames reduces computational load while maintaining sufficient accuracy for downstream reasoning operations.
Solution Approach 2:
The system extracts only the necessary visual features and object states from selected frames, rather than processing complete frame data. This extraction approach isolates the critical information needed for object-level reasoning, reducing the computational burden while preserving reasoning accuracy.
2Ease of operation
If human eye gaze data is used for scene analysis, then attention reasoning is improved, but accuracy deteriorates due to irrelevant regions
Solution Approach 1:
The system applies different processing qualities to different regions of the scene based on their relevance to driving tasks. Critical regions receive higher analysis quality while irrelevant regions (where human gaze may fall) are processed with lower quality or skipped entirely, improving overall accuracy without sacrificing attention reasoning capabilities.
Solution Approach 2:
The system introduces an intermediary filtering mechanism that mediates between human gaze data and scene analysis. This intermediary layer filters out gaze-directed regions that are irrelevant to driving tasks, allowing the system to benefit from attention reasoning while avoiding the accuracy degradation caused by irrelevant visual information.
3Measurement precision
If goal information is required for object-level reasoning, then reasoning accuracy is improved, but system complexity increases
Solution Approach 1:
The system enables object-level reasoning to operate with minimal external goal information by allowing the reasoning module to autonomously infer relevant goals from the current scene context and object states. This self-service capability reduces the need for complex external goal specification while maintaining reasoning accuracy.
Solution Approach 2:
Instead of requiring goal information as input to drive object-level reasoning, the system inverts the approach by having the reasoning process generate relevant goals from observed object interactions and scene dynamics. This inversion reduces system complexity by eliminating the need for complex goal specification mechanisms while preserving reasoning accuracy.
Data Source
AI summary
A system and method for providing object-level driver attention reasoning with a graph convolution network that include receiving image data associated with a plurality of image clips of a surrounding environment of a vehicle and determining anchor objectness scores and anchor importance scores associated with relevant objects included within the plurality of image clips. The system and method also include analyzing the anchor objectness scores and anchor importance scores associated with relevant objects and determining top relevant objects with respect to an operation of the vehicle. The system and method further include passing object node features and edges of an interaction graph through the graph convolution network to update features of each object node through interaction with other object nodes and determining importance scores for the top relevant objects.