Pedestrian Trajectory Prediction Using Waypoints and Correction Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pedestrian trajectory prediction methods, such as Social-STGCNN, fail to accurately reflect social norms and intermediate events in long-term predictions, leading to inaccurate trajectory predictions.
Innovation Solution
A pedestrian trajectory prediction apparatus that includes a waypoint learning unit to learn waypoints from a source to a destination and a corrected trajectory learning unit to learn a correction vector for improving trajectory accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Social-STGCNN uses graph convolutional network to predict pedestrian trajectory, then the prediction can be performed using spatio-temporal graph information, but the prediction accuracy is insufficient because it does not reflect social norms and intermediate events
Solution Approach 1:
The trajectory prediction is segmented into multiple waypoints along the path from source to destination, rather than predicting the entire trajectory as a single continuous path. This allows the model to capture intermediate events and social norms at specific points along the trajectory, improving both accuracy and adaptability to social contexts.
Solution Approach 2:
Waypoints are introduced as intermediary elements between the source and destination. These waypoints serve as intermediate targets that encode social norms and intermediate events, allowing the prediction model to incorporate social context without directly modifying the final trajectory prediction mechanism.
2Duration of action of moving object
If the prediction model considers long-term trajectory, then it can capture overall movement pattern, but it fails to consider events occurring in intermediate stages
Solution Approach 1:
The long-term trajectory is divided into multiple segments connected by waypoints. Each waypoint represents an intermediate stage where specific events or social norms can be captured, allowing the model to maintain long-term prediction capability while preserving information about intermediate events that would otherwise be lost in a continuous long-term prediction approach.
Data Source
AI summary
A pedestrian trajectory prediction apparatus includes a waypoint learning unit configured to learn waypoints on a trajectory for a pedestrian from a source to a destination and a corrected trajectory learning unit configured to learn a correction vector for correcting a trajectory connecting the source and the destination.