Vehicle Trajectory Prediction With Outlier Filtering and Model Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting a target vehicle's trajectory in a vehicle environment are not robust enough to handle outliers caused by random or systematic errors in sensor data, particularly affecting position, speed, yaw, and yaw rate predictions, and struggle to provide reliable estimates up to five seconds in advance.
Innovation Solution
A method that captures vehicle states and road markings using a camera-based device, preprocesses data to remove outliers, calculates both physical and driver-behavior-based trajectories, and combines them using an optimization algorithm, incorporating noise removal techniques like the alpha-beta filter and RANSAC, and utilizes a modified CYRA model for trajectory estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If physical models or driving-maneuver-based models are used for trajectory prediction, then the prediction can be made, but the models must be adapted in a relatively complicated manner
Solution Approach 1:
The patent combines multiple trajectory prediction models (physical model and driver behavior model) into a unified framework that integrates their outputs through weighting factors. This merging approach allows the system to leverage the strengths of each model while reducing the complexity of adapting individual models, as they can be applied in parallel rather than requiring complex sequential adaptation.
2Ease of manufacture
If camera-based capture device data is used directly for prediction, then the process is simple, but outliers caused by random or systematic errors in measured states significantly affect prediction accuracy
Solution Approach 1:
The patent introduces an intermediary processing step that calculates multiple candidate trajectories from the raw camera data and then combines them using weighting factors. This intermediary layer acts as a buffer between the noisy sensor data and the final prediction, filtering out outliers while maintaining the simplicity of using camera-based capture devices.
Solution Approach 2:
The system uses feedback mechanisms where the predicted trajectories are continuously refined by comparing multiple candidate trajectories and adjusting weighting factors based on their plausibility. This feedback loop allows the system to correct for systematic errors and random outliers in the camera data without requiring complex preprocessing.
3Reliability
If multiple trajectory models are combined to improve prediction robustness, then prediction accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent applies partial action by using a limited number of candidate trajectories (typically 3-5) generated from the raw sensor data, rather than processing all possible trajectory variations. This partial approach provides sufficient robustness against outliers while keeping computational complexity manageable for real-time processing.
Solution Approach 2:
The system dynamically adjusts weighting factors for different trajectory models based on the specific driving situation and the plausibility of each candidate trajectory. This parameter adaptation allows the system to emphasize more reliable models in given contexts while reducing computational effort on less plausible trajectories, thereby balancing robustness and complexity.
Data Source
AI summary
A method for predicting a trajectory of a target vehicle in an environment of a vehicle. The method includes the steps of a) capturing states of the target vehicle, capturing states of further vehicle objects in the environment of the vehicle and capturing road markings by a camera-based capture device; b) preprocessing the data obtained in step a), wherein outliers are removed and missing states are calculated; c) calculating an estimated trajectory by a physical model on the basis of the data preprocessed in step b); d) calculating a driver-behavior-based trajectory on the basis of the data preprocessed in step b); and e) combining the trajectories calculated in steps c) and d) to form a predicted trajectory of the target vehicle.