Vehicle Path Prediction Using Indicator Activation Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing driver assistance systems struggle to accurately predict the future path of a vehicle and the driver's intentions, particularly in complex scenarios like lane changes and turns, leading to potential collisions due to limitations in recognizing the timing and reliability of maneuvers, and the overlap of driver behaviors.
Innovation Solution
A method that evaluates the probability of a vehicle's future path by using the current position, direction indicator operations, and parameterizing a model to determine the occurrence measure of the direction indicator's activation, allowing for improved prediction of driver intentions and maneuvers, including lane changes and turns, with reduced development and validation effort.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple classifiers are trained for each maneuver type to predict driver intentions, then the probability of detecting specific maneuvers can be estimated, but the reliability decreases when several maneuvers occur in succession and driver behaviors overlap
Solution Approach 1:
The patent segments the prediction task by introducing a hierarchical structure with a first classifier for lane change detection and a second classifier for turn detection. This segmentation allows the system to handle sequential maneuvers by processing them in stages, where the first classifier identifies lane change intentions and the second classifier subsequently identifies turn intentions, thereby improving reliability in sequential maneuver scenarios
Solution Approach 2:
The patent applies preliminary action by using the first classifier to detect lane change intentions before the actual lane change occurs. This early detection allows the system to prepare for subsequent maneuver detection, enabling the second classifier to more accurately predict turns that may follow the lane change, thus improving overall prediction reliability
2Device complexity
If a fixed set of features is used in the classifier to estimate maneuver probability, then the classification process is simplified, but the accuracy decreases in complex scenarios like lane changes and turns
Solution Approach 1:
The patent applies dynamics by making the feature set adaptive rather than fixed. The system dynamically selects and weights features based on the current driving context and detected maneuver type. For example, when a lane change is detected, the system activates features relevant to lane change behavior, and when a turn is detected, it switches to features relevant to turning behavior, thereby maintaining high accuracy across different scenarios without excessive complexity
Solution Approach 2:
The patent changes parameters by adjusting the feature weights and selection criteria based on the maneuver type and driving context. The system modifies the importance of different features (such as steering angle, lateral position, acceleration) depending on whether a lane change or turn is being predicted, allowing the classifier to adapt to complex scenarios while maintaining a manageable system structure
3Measurement precision
If the system waits for the driver to complete mirror checks and shoulder checks before predicting intentions, then the prediction timing is more accurate, but the reaction time is delayed
Solution Approach 1:
The patent applies preliminary action by predicting driver intentions before the driver completes all traditional safety checks. The classifier system analyzes subtle cues from vehicle dynamics and driver behavior to predict lane changes and turns in advance, allowing the driver assistance system to prepare warnings or interventions before the driver fully commits to the maneuver, thereby reducing reaction time while maintaining acceptable accuracy
Solution Approach 2:
The patent uses dynamics by continuously updating the prediction as the driver progresses through their decision-making process. The system adjusts its prediction confidence and timing based on real-time changes in vehicle dynamics and driver behavior, allowing it to optimize the balance between early detection and accurate intention recognition throughout the maneuver sequence
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
The invention relates to a method for predicting a future driving path of a vehicle, comprising the following steps: the actual position of the vehicle, in particular the actual distance between the geographical position of the vehicle and a reference position is used; a possible future driving path of a vehicle is used; information relating to the activation of the indicator is used, said information representing whether the indicator is active with respect to the actual position and optionally in which position, in particular at which distance from the reference position, and in which direction was the indicator last activated; a model for the information relating to the indicator according to the possible future driving path is parameterised, said parameterised model for information relating to the activation of the indicator and the actual position, in particular the actual distance, provides the number of times the indicator is activated, that is a numerical value for the probability that the information relating to the activation occurs on the driving path; the number of activations of the indicator using the parameterised model is determined.