Traffic Participant Tracking With Retroactive Trajectory Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for analyzing traffic scenes in automated driving systems either lack the ability to incorporate past tracking information effectively, leading to tracking errors and inaccuracies in predicting future trajectories, or they persist tracking errors due to sequential processing of detection, tracking, and prediction.
Innovation Solution
A computer-implemented method that aggregates scene-specific information at a current point in time to reconstruct past track profiles and predict future trajectories, integrating tracking and prediction while using a deep learning architecture to generate latent features for improved accuracy and robustness, thereby reducing the impact of measurement outliers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional methods use sequential processing (detection first, then tracking, then prediction), then the process structure is simple and easy to implement, but tracking errors persist and cannot be rectified in subsequent steps
Solution Approach 1:
The patent merges detection, tracking, and prediction into a single unified neural network model that processes all tasks simultaneously. The model takes current sensor data and historical track profiles as input and outputs both detection results and updated track profiles in one forward pass, eliminating the sequential processing pipeline and allowing error correction across all functions.
Solution Approach 2:
The patent reconstructs past track profiles retroactively using current perception results before final prediction occurs. By reprocessing historical sensor data with the current model state, the system corrects accumulated tracking errors from the past, ensuring that the prediction is based on accurate historical information rather than propagating old errors.
2Reliability
If methods use online tracking with traditional trackers, then past information can be used for prediction, but tracking errors accumulate and persist in each further tracking step
Solution Approach 1:
The unified model incorporates feedback by using the current detection result to retroactively reconstruct and correct the entire track profile history. The model compares current perception with historical track data and adjusts the reconstructed profile to eliminate discrepancies, ensuring that corrected information feeds into the prediction process rather than allowing errors to propagate forward.
Solution Approach 2:
Before generating the final prediction, the system performs preliminary reconstruction of the track profile by reprocessing past sensor data through the current model. This preliminary action corrects historical tracking errors by aligning past observations with the current understanding of object states, ensuring clean input for the prediction stage.
3Ease of manufacture
If methods without online tracking are used, then the implementation is simpler, but only current perception results are available for prediction without additional track profile information
Solution Approach 1:
The system performs preliminary reconstruction of complete track profiles by reprocessing historical sensor data through the unified neural network. This reconstruction process generates accurate historical state information (positions, velocities, orientations) that would otherwise be lost, providing rich temporal context for prediction while maintaining a unified model architecture that remains relatively simple to implement.
Data Source
AI summary
A computer-implemented method analyzing the behavior of a participant in a traffic scene. In the method, the participant is detected on the basis of aggregated scene-specific information in the traffic scene. At least one past track profile for the participant is reconstructed on the basis of the scene-specific information aggregated at a current point in time, by generating perception results for a sequence of points in time in the past.