Vehicle Target Selection Using Trajectory Similarity Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current target selection methods for vehicles with advanced driver assistance systems (ADAS) or autonomous driving are computationally intensive and lack an efficient approach to differentiate between road users based on their impact on the vehicle's trajectory, relying on rule-based systems that are not adaptable to various driving behaviors.
Innovation Solution
A computer-implemented method that predicts vehicle and road user trajectories using dynamic and static information, performs similarity comparisons to determine potential targets, and adjusts relevance thresholds to reduce computational load and improve target selection accuracy, employing machine-learning algorithms like convolutional neural networks and recurrent neural networks for context-aware scene prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional rule-based target selection is used, then target selection can be performed for different applications, but computational resources are excessively consumed
Solution Approach 1:
The patent changes the parameter of target selection from rule-based to similarity-based comparison. It introduces a similarity threshold parameter that dynamically filters potential targets by comparing their trajectories with the ego-vehicle's trajectory, thereby reducing the number of targets requiring further processing and lowering computational power consumption while maintaining selection accuracy
Solution Approach 2:
The patent performs preliminary trajectory prediction for all detected road users before applying the similarity comparison filter. This preliminary action of predicting trajectories in advance allows the system to efficiently identify and eliminate irrelevant targets using the similarity threshold, reducing the computational load on subsequent target selection processes
2Adaptability or versatility
If traditional rule-based target selection is used, then target selection can be performed for different applications, but the system lacks adaptability to various driving behaviors
Solution Approach 1:
The patent replaces rigid rule-based parameters with a flexible similarity threshold parameter. This parameter can be dynamically adjusted based on driving scenarios, road user behaviors, and environmental conditions, enabling the system to adapt to various driving behaviors without requiring complex rule sets for each scenario
Solution Approach 2:
The patent substitutes the mechanical rule-based system with a data-driven similarity comparison approach. Instead of following predefined rules for different driving scenarios, the system uses trajectory similarity metrics to automatically identify relevant targets, reducing system complexity while improving adaptability to unseen driving behaviors
3Reliability
If all detected road users are considered as potential targets, then no targets are missed, but computational load increases significantly
Solution Approach 1:
The patent introduces a similarity threshold parameter that filters potential targets based on their trajectory similarity to the ego-vehicle. This parameter change allows the system to quickly eliminate road users with dissimilar trajectories, maintaining reliability by preserving all potentially relevant targets while significantly improving processing speed through early filtering
Solution Approach 2:
The patent applies partial action by performing full trajectory analysis only on road users that pass the similarity threshold filter. Instead of analyzing all detected road users in detail, the system performs a lightweight similarity check first, then applies comprehensive analysis only to the subset of potentially relevant targets, optimizing the balance between completeness and processing efficiency
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Computer implemented method for target selection in the vicinity of a vehicle, comprising obtaining vehicle state information, the vehicle state information comprising dynamic information regarding the vehicle, predicting a first trajectory of the vehicle based on the vehicle state information for a first prediction time horizon, detecting road users in the vicinity of the vehicle, determining state information from the detected road users, the state information comprising dynamic information regarding the road users, predicting a second trajectory of the vehicle based on the vehicle state information and the road users state information for the first prediction time horizon and performing a first similarity comparison of the first predicted trajectory and the second predicted trajectory of the vehicle to determine whether the detected road users are a potential target of the vehicle for the first prediction time horizon.