Pedestrian Action Prediction Using Spatiotemporal Road User Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Predicting pedestrian actions in autonomous driving systems is computationally intensive due to the need to model complex interactions with the surrounding environment, which are dependent on multiple environmental factors and spatiotemporal context.
Innovation Solution
A method involving spatiotemporal graph modeling of road users, clustering, and using a trained graphical convolutional neural network to predict pedestrian actions, incorporating relative importance weights based on distance and directional location to enhance prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional methods are used to model pedestrian interactions with surrounding environment, then prediction accuracy can be maintained, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent segments the complex environment into discrete road user objects, each represented as a node in a graph structure. This segmentation allows the system to process individual road users separately rather than modeling the entire environment as a single complex system, reducing computational complexity while maintaining prediction accuracy through localized interaction modeling.
Solution Approach 2:
The patent transforms the continuous environmental data into discrete graph-based representations with specific parameters (node features, edge weights, spatial relationships). This parameter transformation converts complex continuous modeling into structured discrete computations, improving processing efficiency while preserving essential interaction information for accurate prediction.
2Measurement precision
If comprehensive environmental factors and spatiotemporal context are considered in pedestrian action prediction, then prediction accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary encoding of road user features and pre-computation of spatial relationships during the graph construction phase. By preparing these data structures in advance before the actual prediction task, the system reduces the computational burden during real-time processing, achieving both high accuracy and efficient processing time.
Solution Approach 2:
The patent implements dynamic graph updates that adapt to changing environmental conditions. The graph structure and edge weights are updated based on current spatiotemporal context, allowing the system to focus computational resources on relevant interactions rather than processing all environmental factors uniformly, thus improving processing efficiency while maintaining prediction accuracy.
3Measurement precision
If relative importance weights based on distance and directional location are incorporated, then prediction accuracy improves, but model complexity increases
Solution Approach 1:
The patent applies local quality by assigning different importance weights to different road users based on their specific spatial relationships with the target pedestrian. Instead of using a uniform model for all interactions, the system customizes the weighting scheme locally for each road user based on distance and directional location, improving prediction accuracy through localized optimization.
Solution Approach 2:
The patent implements partial action by selectively computing importance weights only for relevant road users within a certain spatial range, rather than computing weights for all possible objects in the environment. This selective approach reduces model complexity and computational overhead while maintaining prediction accuracy for the most influential road users.
Data Source
AI summary
The present disclosure relates to methods and systems for spatiotemporal graph modelling of road users in observed frames of an environment in which an autonomous vehicle operates (i.e. a traffic scene), clustering of the road users into categories, and providing the spatiotemporal graph to a trained graphical convolutional neural network (GNN) to predict a future pedestrian action. The future pedestrian action can be: one of the pedestrian will cross a road and the pedestrian will not cross the road. The spatiotemporal graph includes a better understanding of the observed frames (i.e. traffic scene).